Hewlett Packard 8640B Signal Generator

See an explan of the equipment

 

 This is what the seller termed a "barn find" and the description isn't bad. I've had a few pieces of equipment that have suffered from being in a damp garage for donkeys' years and this is pretty typical. The main problem, at least exterior-wise is aluminium corrosion and the worst aspect of this is damage to lettering on dials. The front panel isn't too bad as it's mainly pin-hole corrosion that should clean off.

Another problem though, that can be serious, is rust affecting things like grub screws tapped into brass and expansion/contraction of fittings using plastic/metal combinations especially if plastic goes brittle and disintegrates.

After a lot of trouble I finally removed the knobs using 0.05 inch hex grub screws, but the larger screws are not yet removed. Only one knob looks really bad and may need a printed paper scale replacement. I find that it's only after investment has been made that one discovers life-threatening problems.. whether it be test equipment or cars! Still at £45 the gamble here is not too risky.

Below, the 8640B with the top cover detached and the upper panel removed and the lower panel partially cleaned up.

 

 The seller said he'd plugged it in and nothing had happened, but before I tried to power it up myself I thought I'd check the mains setting. Being made in the USA it will have a 120/240 volt setting. This is to be found hidden away, to be seen only in good lighting or using a torch. Adjacent to the IEC connector you'll see a fuse and a small lever for removing it. Remove the fuse and you'll see a tiny printed circuit board and on this in gold-plating are numbers, which in my case included "240". Apparently you can pull out the board using a special tool and turn it over to be then equipped for 120 volt operation.

I plugged in a kettle lead and pushed the on/off button. I should mention that I'd already checked and seen about 20 ohms across the live/neutral IEC connector pins so was fairly confident of something interesting happening. There was a sort of coughing noise and the fan came on... it then intermittently went off for a fraction of a second before coming back on so something is not quite right. The front-panel display lit with various similar red characters but flashing at a regular rate... usually with other things this means that there's a (not too serious) problem preventing normal operation. This might be an under or over voltage somewhere in the power supply, but until I've studied the user manual I can't yet tell.

After gazing at the display for about a minute hoping it would suddenly say 500MHz there were cracking noises like a firework and a little flame appeared at the rear of the chassis, followed by volumes of smoke. Fortunately, this is not an unknown occurrence here, so I just waggled the mains lead free and what might have been panic.. went away. The culprit was a mains suppression condenser. One of those chubby semi-transparent gold coloured things.

Exactly the same thing happened when I plugged in my BBC B computer after a score of years being hidden away (and on lots of circuit boards for lifts.. my day job is repairing such things).

I have several 8640B documents downloaded from the Net and in one I (eventually) found the mains input circuit diagram revealing C6 as the failed condenser. This is wired across the rear of a filter and before the mains switch so the designers must have put their money on the mains fuse to prevent a conflagration?

 
 

 Why these things fail, I don't know, but they do fairly regularly. Maybe it's something to do with their rating as this example says "250VAC" on the side.

As most electrical engineers know, the UK does not have the widely published 230v mains but 240v mains which can rise a lot higher than 250 volts. Of course it's the tolerances that matter not the actual number.

I just cut the legs and removed it for now.

 

 

Below, the initial bad components.

 

 

 This board (covered by Service Sheet A22) whose circuit diagram is shown above as "A12 rectifier assembly 08640-60003", is located between the mains transformer and the set of large smoothing capacitors and carries five bridge rectifiers each using four identical Motorola diodes coded 14-18 with 2 legs, in cans not unlike TO5 (cathode to can). Three were open circuit (unusual for a diode as these generally fail short-circuit) and two measured with Vf about 5 to 10% higher than the remainder of the good diodes using a diode tester. My guess is that these two had been stressed by failure of the other diodes and likely to be unreliable.

The diodes are 1901-0418 (Motorola SR1846-12) probably rated at 1200 volts (unnecessarily) and mechanically equivalent to the 1N4999 rated at 3A and 200 volts. The HP documentation mentions both 1.5A and 3A so take your pick but the higher current rating is better.

Modern replacements could be 1N5402, BY255 or BYW95.

The last being in a SOD64 package which can fit better as the board sits very close to the smoothing capacitors. However the hole sizes for the diodes vary with CR9/CR11 being a lot smaller than typical 1.3mm 3A diode legs with SOD64 having 1.35mm legs.

 
   

 I decided to use a set of 1N5408 because I have several of these. The anode pin sits centrally in the original diode so the cathode of the new diodes is bent over as close as possible to the body to minimise height. Before bending the cathode legs I filed these down to fit the mounting holes. The two new diodes on the left however are mounted differently because the mounting hole was only around 0.75mm diameter. The cathodes are soldered directly to the gold plated areas. In one case I added a through wire to carry current rather than relying on the hole plating.

The board is tricky to fit back in place because of the limited access but having done this I powered the thing up. This time the fan didn't come on, which it had done previously (although intermittently) and since switching the counter mode lock to OFF the display came on without flashing, however it read all zeroes rather than displaying a frequency. Pressing a few buttons (those that are no longer seized) caused lamps to come on. Pressing the lock button caused the display to flash so it seems the equipment is at least alive. Some switches are still seized including the RF ON/OFF and the leftmost modulation switch which has a broken lever (and suggests a serious fault).

Why does the display show all zeroes? Several possibilities including the obvious one that the oscillator isn't working or that the various frequency dividers are faulty. There are other reasons, more likely.. a power supply problem or a wrong switch setting. This equipment is actually two instruments in one.. a signal generator AND a frequency counter (my Marconi TF2002 for example, because the tuning arrangement is just a linear scale aided by a crystal reference requires an external frequency counter before one can see exactly what the output is) so obviously all zeroes may indicate the counter circuitry isn't working. There are some basic tests to enable one to work out the reason for the problem, but initially I'll just check the power supply voltages, in particular the 5 volt rails. After the power had been applied I had noticed one of the set of red LEDs was dimmer than the others... so armed with a voltmeter I again plugged in the mains lead.

I was just about to press the on/off switch when there was a flash and a bang and smoke so I quickly removed the mains cable. The 1.25A fuse was very black. That meant a catastrophic fault. Looking at the circuit diagram for the second time I realised the problem must be in the mains filter area because this is before the on/off switch (see below). The mains filter is integral with the IEC connector assembly which is a fairly small, sealed box fitted into a vertical slot in the rear panel.

 

 

 Access to the filter is awkward. It's secured by a pair of spring steel plates that grip the inside of the rear panel, and to remove the assembly you have to press down first on the front edges of the lower plate whilst levering the lower edge of the mains connector. Ideally the power supply motherboard should be detached but, by using a couple of thin screwdrivers through convenient apertures, the lower edge can be freed. Once this has unclipped the same method is used to release the easier top clip and the assembly pulled back until it's free from the rear panel. A really tight fit.

The cable prevents the filter assembly from being pulled out completely, but sufficient to get a drill in position to remove the four pop-rivets holding the external metal shield.

Once the top of the shield has been bent back you can see a ferrite filter coil wrapped in yellow tape. Unsure of the actual failure I unwrapped the coil but found it was OK. In the front edge of the filter are two long orange/brown coloured feed-through capacitors and tracking could be seen near one of these. I pulled off the wires (which go to the on/off switch) and broke away the capacitors. One was clean but the second was blackened.

My first thought was to replace the whole assembly with a small metal plate holding an IEC connector, but repair was better because the filter assembly also includes mains voltage selection (with the tiny PC board) and all the transformer primary wiring (below left). I found a couple of ceramic feed-throughs which I soldered in place of the originals and added a pair of 400VAC capacitors as filters. I could only use one self tapping screw in place of the old pop-rivets because of the tight fit of the filter so I used solder to hold the screen together. The cut red wires went to C6.

Below, connections to the rear of the IEC connector.

 

 

 

 

 Above.. the repaired filter assembly being forced back in place. It's an interference fit. A new 1A fuse.

 Above... The filter back in place with the mains switch wires soldered to the live sides of the capacitors.

 The next task was to continue testing as I'd intended immediately before the mains filter failed. Before plugging in and switching on, I removed the series pass power transistor fitted underneath the chassis as this is reported to develop poor contacts with its mounting socket. Sure enough the 2N3055 legs had traces of corrosion and after cleaning I refitted the transistor, plugged in and switched on. The red power supply LEDs all lit and the display came on with all zeroes as before. I checked the various voltages at the test points on the plug-in boards and all looked reasonable.

I then proceeded to press the various buttons and switches. Some carried no markings and as the top front panel was detached I hadn't checked to see what these were for. To my surprise, the lower left button of the group of six in the centre resulted in a display. Checking the front panel revealed the answer. One can use the frequency counter to read either the internal signal frequency OR the frequency of a signal at the BNC counter input. The reading could be varied with the tuning knob from 3352 to 4168 although I can't say in what position is the range switch because the knob is detached and the spindle doesn't want to turn easily. The HP8640B uses lots of plastic gears which over the years have degraded. My example has several gear wheels with cracks in their hubs and in a couple of cases extending to the teeth.

As gear problems are so common with this series of equipments there are several repair solutions which I need to investigate further. There's also the tuning problem. The tuning knob rotates through only around 450 degrees... from a solid-feeling end-stop to a springy stop so maybe a damaged gearwheel is the root cause. The tuning mechanism not only drives the oscillator cavity but also includes links to a couple of ancillary parts so maybe one of these hitting its endstop is the reason for the fault?

 

I removed the cover from the pcb carrying the display because that circuit board also carries the set of six push buttons and the x10 is completely frozen. Detaching the pcb was slightly puzzling until I noticed you don't pull the board outwards but lift the riser into which it's plugged. There was discolouration around the seized switch but after lots of switch cleaner and gentle persuasion the thing started to work and after lots of presses worked perfectly. With that working I found the x10 and x100 buttons did exactly what they should.

Next the meter. This records precisely zero reading with any button or switch position except for a tiny reading in one of the FM settings.. Could the RF output be knackered? Lack of voltage depends on not only the condition of the RF stages, but also a fault in the area from where the meter circuit gets its information I suppose. A couple of signs make me think the situation is bad. Firstly the digital readout seems stuck in the range 3352 to 4168. If this is in KHz it equates to the 80m amateur band, and if a transceiver has been tested... has the user inadvertently switched to transmit and blown up the output circuitry of the signal generator? This example appears to have little or no protection, and rumour has it that lots of examples of the HP8640B are short of an astronomically priced RF power hybrid for said reason.

I decided to bite the bullet and see if there's any RF output so carried the thing to the workshop and connected it to a scope and turned it on. The display went a solid yellow. I pushed the Auto button and a nice sinewave was revealed. "Measure" gave me around 4.1MHz at over 5 volts RMS. I twiddled the attenuator and the trace responded so all is well.. the RF output is fine. Now I have to work out why the tuning range is foreshortened and why the meter fails to work... and fix the AM switch which is seized.
 

I made a fair bit of progress without achieving much. It seems, because the HP8640B was so popular it was manufactured for ages, and over the years various improvements were made. You have to keep your design engineers employed after all... and lots of cash must have rolled in so wages weren't a problem (especially underpaid engineers). The various electronic modules look much the same but careful inspection revealed some significant differences. I'm trying to track down the lack of any meter readings.. that goes for AM or FM modulation and RF output. After printing out circuits showing the path from the RF amplifier to the meter and looking for problems I found that reality and the documents didn't line up so I looked in a different manual and again, printed out the same path. This is much better as it looks almost right. For example, the actual meter is connected to an op-amp in one version of the manual and a transistor in the second. Give the job to an engineer and he'll come up with a much more complicated solution than necessary so changing from a transistor to an op-amp was an excellent idea.. not to mention adding a raft of auto-ranging circuitry. "Keep it simple", sounds great but won't keep your team of engineers busy.

Using the second manual I can now follow the RF meter input backwards from the meter board to its local motherboard, then a small riser connecting to a second, smaller motherboard where the AGC board resides. At this point the track disappears. I expected it to make its way to the AGC board on which are located some processing circuits dealing with a rectified DC level from the RF output board, but does the track go via the RF board or is there a break or perhaps a small diode under the motherboard which cannot be detached without loads of effort?

At this point I wondered if the AGC board was fitted the right way around. None of the plug in boards carries a polarising key, but instead is fitted with coloured levers which match coloured guides, but in some cases the plastic handles had gone brittle.. hence no handles and therefore no immediate clue to orientation. I looked further and noticed numbers on the motherboard indicating the plug numbering. To my surprise the AGC board was the wrong way round. Maybe this explains the lack of a meter reading? I switched it around but still no meter reading. There's a strong possibility the reversal has blown up a transistor or something vital, but checking the circuitry showed damage is unlikely except perhaps a transistor or two or maybe an element in a 74 series package (7406 and 7402). However, I don't really see that any damage could have affected the metering circuits. The correct circuit diagram (which I can't find foc on the Net) for this specific equipment should reveal why there's a discontinuity in the meter circuitry. I got bored so moved onto the tuning problem...

 

 The good news is that I found the cause of limited tuning and it's got nothing to do with the oscillator cavity as such. In fact I've seen the problem before in my DST100 receiver and it's almost exactly the same problem for a different solution to the mechanical design problem of counting rotations of a tuning mechanism. Basically you need to achieve acceptable mechanical tolerancing down to a single knob rotation to a linear shift in the mechanism. Say you need to count 10 turns.. then you can allocate one tenth of an inch for a mechanism depth of say one inch, plus an allowance for a (slightly wobbly) gearbox. So if a depth of 15mm is available you can allow 5mm for a gearbox or frame and 1mm for each turn of the tuning knob. During manufacturing you can plan to adjust things so that the mechanism works perfectly, but you're dealing with inaccuracies such as operator setting up in manufacture and manufacturing tolerances in the different parts. Given an expected equipment life of say 20 years you'll get wear and tear which will eventually result in the mechanism failing.

So what exactly went wrong? The plastic disks each have an oblong pip with square corners and as wear increases the position of the pips will move from optimum to only just working. Then, given a spot of rough handling, the square corner of a pip might get worn and, instead of locking against an end stop, slightly more rotational force than normal will allow an extra turn of the knob. This is what had happened and at this point I'll refer to the cavity tuner to now be in no-mans-land. The piston has moved beyond its design stroke and the indicated frequency will be higher or lower than expected, depending on whether the forced turn was clockwise or anti-clockwise. Now, if the knob is turned back, instead of freely moving the piston all the way back, the wrong side of the next pip will hit the endstop. You cannot exit from no-mans-land and the tuning range will be about a tenth of the normal range with the frequency coverage extending out-of- range. In my case 3352 to 4168KHz, being part of the 2-4MHz range because the piston has pushed too far into the cavity. Is the fact that this range of frequencies covers the 80m amateur band a clue to the demise of this example of the HP8640B?

Anyway...no-mans-land can be exited, at least temporarily, by gently pressing on the first locking pip (below left), then the next and so on.. but turn the knob the other way and the worn pip (below right), instead of waiting for several to back up, will slip past the stop and you're back in no-mans land... The solution is either to re-align the mechanism or to repair the worn pip..... below, you can see about 1mm is available for adjustment to overcome the damage from the peg wearing away more of the leading edge of the last pip. Possibly a small sliver could be cut from the pip to square its leading edge, but access is tricky?

 

 

 

 Returning to the lack of a meter reading and armed with prints from the three manuals I'd downloaded (none of which is quite right) I was determined to track down the meter fault. Part of the problem is understanding the way the diagrams are coded and another is having to keep turning 25kgm of chassis upside down and back again. I'll summarise what I found over the period of a day interspersed with spells of gardening. Working backwards from the meter. Two wires clip to pins on the metering motherboard where tracks go to the meter pcb. Here, various switches and op amps develop a sensible meter current together with an indication on lamps as to which meter scale is relevant. The meter circuit is driven from a voltage arriving on a meandering printed circuit track heading forward from the rear of the equipment. Having checked that no damage would follow, I connected an AA cell negative to the chassis and positive to this track. The meter swung to full scale and the lowest lamp came on. The meter pcb seems to be OK.

The long track, visible from the top of the chassis, goes to a short riser which connects to another motherboard underneath the chassis. Here it goes to another riser to yet another motherboard on which the AGC board is fitted. This is accessed, like the meter pcb, from the top of the chassis and carries an op amp which handles the voltage from the circuit used for detecting the RF. Again, checking that no damage could occur, I connected the AA cell, but this time to the input of the AGC pcb. The meter failed to move, so I reversed the polarity, feeding a negative 1.5 volts to the input.. and the meter swung to two-thirds scale. The AGC board is working. Next, I traced the track backwards from the input to the AGC pcb. A track goes across the motherboard back to the riser, but from there, instead of more track, there was a bunch of wires heading off to the rear of the attenuator switch where a rotary printed circuit switch (I refer to this type as a mode switch because these were commonly termed as such in VCRs for selecting circuit changes associated with record and play). There's also a potentiometer whose shaft goes through the attenuator case to the centre of the attenuator switch. The mode switch meter circuit has two wires and a pair of resistors used to interface between the AGC input and the RF amplifier. The opposite side of the riser, to that connecting to the AGC motherboard, connects to the RF power motherboard which is sandwiched between pcbs in the RF box. On the underside of this box, under a lid secured by a set of screws, is the pcb carrying the RF power output circuit. Included on this pcb is a circuit for processing the rectified RF signal and its output voltage is the one that passes through the attenuator mode switch to the AGC pcb.

The voltage at the output is an unvarying 0.186 volts, so there must be a fault in the rectifying and switching circuit or in one or more of the several signals from the front panel controls, which includes the RF on/off switch. To investigate exactly what's happening to the meter signal I've shown below its path starting from the RF power pcb and ending at the meter.

 

 

 

 Left, the RF output pcb, 26A1. The RF power hybrid circuit carries an RF detector diode and filter which feeds circuitry (lower left) where it's processed in accordance with signals from the various equipment controls (the group of coloured wires) and outputs via the small transistor top right of the pcb (Q1). The tiny pot above the transistor sets the DC level. From the wiper of this pot it passes via the grey wire to the motherboard P26A6, below the pcb, and from there to the connector and the riser at the side of the RF box, below left, connecting to the PSU motherboard.

 

 Below, before heading off from the meter amplifiers the DC signal passes through the dark yellow wire, via a mode switch coupled to the main RF attenuator. Depending on the attenuator setting one of two resistors is inserted into the path before it returns to the riser via the blue wire. Later, I found that one of these resistors defines the normal power output of 10dBm and the second a higher value of 19dBm used for the +20dB attenuator range.

 Top right you can see the corner of the PSU motherboard where it connects to the RF power riser. Connected to this riser are several wires going to the mode switch coupled to the RF Attenuator (bottom left).

 

 

On the right is the rear of the main cavity oscillator assembly

 

 
 This is 26A6, the motherboard in the RF box. The signal from the attenuator mode switch passes to the AGC amplifier pcb. You can just see the socket for the AGC amplifier pcb at the lower edge of the picture. Directly underneath this motherboard is the RF output pcb A26A1.
 

 

 

 Above is PSU motherboard and on its left, against the RF power box are the connections to the riser connecting through to the meter motherboard.

  Above, the AGC amplifier pcb (lower) and the AM Offset pcb, 26A2 which handles the interface between the front panel controls and the meter switching on P26A1. The meter signal enters the AGC pcb 26A4 at Pin 7 from motherboard 26A6 where it enters above left from the riser carrying the wiring from the attenuator mode switch.

The riser connects the meter signal from P26A6 to the power supply motherboard.
 

 

 

 Above, the signal enters the meter motherboard at Pin 30 from the riser.

Right, the amplifier on the meter pcb is fed by the rightmost gold track at Pin 9, and outputs back to the Meter motherboard to the red and white wires connecting directly to the meter. The Meter pcb carries three push switches, the lowest for the RF voltage/power level. In addition to the meter voltage circuitry, another circuit evaluates the appropriate meter scale lamp to be lit.

 

 Having worked out the path of the meter signal representing RF output I had already checked the voltages, so re-capping...Clearly the meter was reading zero volts but what was the input to the meter pcb? It was less than 100mV so after checking that no damage would result I connected an AA cell and saw more or less full scale deflection for +1.5 volts. I then repeated the test, but this time at the input to the AGC pcb. No result with 1.5 volts but, again a good meter deflection with the AA cell reversed and putting -1.5 volts into the meter amplifier. So, all things considered both the AGC and meter pcb's are working. What then is the output from the RF power amplifier pcb? It should be something like, at least minus 1.5 volts, but in fact it read +0.186 volts , hence a zero meter deflection.

By now, I'd decided there were too many differences between the free manuals so I bit the bullet and purchased a download for the manual describing my own HP8640B (1625U). Looking at the RF amplifier pcb (see below) there are differences between most of the circuits so maybe there had been problems in earlier models (or of course a problem with 1625U?). A key piece of information is the correct voltage emerging from 26A1, which is around minus 3 volts. Mine, as I said was plus 186mV. The manual tells you to check Q1 and Q2 plus associated circuitry, but of course the fault could be as far back as the basic RF output detector diode or even low RF output.

Sticking to the recommendation, I looked at the emitter of Q1. This was sitting at about 9.5 volts, and the base of Q1 about 9 volts. The emitter of Q1 goes via a 5.1Kohm resistor to +20 volts so the emitter current should be (20-9)volts divided by 5.1Kohm = 2.1mA. The collector of Q1 connects through a 10Kohm resistor plus around 130 ohms, representing the 200 ohm pot plus about 400 ohms from R16 and R17. Ignoring any current drawn through CR6 and any Q1 base current, the voltage at the collector of Q1 should be determined by the emitter current of 2.2mA. The resistance is 10.13Kohm so the voltage is 10.13Kohm x 2.2mA. = 21.2 volts. This sets the collector at around plus 1.2 volts with the pot supplying a max change of only 0.3 volts. Therefore the meter will read zero volts because the voltage at the wiper of the pot should be less than zero to produce a meter deflection. Allowing say a Q1 base current of say 0.25mA the collector will be 10.13Kohm x 1.95ma = 20-19.75 = minus 250mV The measured voltage was +186mV making the base current a bit less. The reason for the low voltage must be something amiss in the circuitry either between the power amplifier RF detector and the base of Q1, otherwise a low RF output.

 

 Working backwards from something said to be correct. . If the test point DET is minus 3 volts (Note: re-reading this later I realised that this figure relates not to +10dBm but to the maximum output where around +19dBm was the target.. a loop based on +10dBm would make DET around minus 1 volt) the collector current of Q1 should be about 17 volts divided by 10.13Kohm = 1.68mA. This is roughly reflected into Q1 emitter current making the emitter voltage about 20-(5.1Kohm x 1.68mA) = 11.4 volts. It's actually 9 volts, so either the rectified voltage from the power amplifier is low (0.55 volts) or this is being reduced by loading from Q6, Q7 or Q3. Otherwise the sample & hold circuit is incorrect and not allowing Q3 source to rise to 3 volts. I noticed the detected voltage doesn't vary when the AGC switch is set to on or off. I imagined the RF output should be held to a precise level so the attenuator settings will be correct so maybe the RF output is stuck without any AGC action, or the AGC circuit isn't working?
 

Next, I'll investigate the AGC system and ensure this is working correctly before moving to the RF amplifiers. AGC is carried out on the A26A4 pcb. Checking the output of the feedback loop, which maintains the RF output into the attenuator at a precise figure, I found this to be about 15 volts. This represents the maximum drive level and will force the RF output amplifier to its maximum output.  This is really to be expected because the AGC input is driven from much the same sensors as the meter drive. In other words, if the detected RF output is too low to drive the meter it will drive the AGC circuit to increase the output until it reaches the correct level. If the AGC loop is inoperative the drive to the RF output stage will be the maximum of 15 volts. We can test the AGC loop quite simply by setting the RF on/off switch to off. This reduces the output from +15 to -1.7 volts so, what then is the detected RF output? I checked the voltage at R22 and found this to be 580mV and this was unchanged whether the RF switch was set to on or off. Could the fault lie in the RF amplifier? This is a hybrid device (used in both the 8640A and 8640B) for which I found the circuit below..

The detector output appears to be fixed at a low level. The HP manual tells me the hybrid amplifier has a gain of 16dB and I've already established a decent RF power level is present. Could the detector be faulty? The detector diode is driven via a 200 ohm resistor with a 33pF shunt capacitor perhaps used to eliminate variations in the self capacitance of the diode as well as reducing the diode current with RF filtering provided by a 100pF capacitor. I soon discovered the circuit below which I understand has been drawn from an examination of the inside of the hybrid.

 

 Looking at the circuit above, one could test the detector diode using a multimeter as R17, R15 and R16 offer a path to ground. In fact the diode circuit measured 1.095 volts one way and infinite the other. To confirm this is about right I connected an OA90 diode in series with a 1.2Kohm resistor and then a GEX54 diode. The reading I got with the former was 1.2 volts and the latter was 1.0 volts with infinite in reverse. Ostensibly the path which includes R15 proves that if RF output is present the detector circuit should work normally. This means that unless the hybrid has a very peculiar fault the problem must be on the 26A1 pcb. Candidates are C4 and Q6, C5 and Q7, or Q3. However, the HP8640B has fooled me a few times before, so thinking a little more... the penny dropped. I measured the voltage at the junction of R22 and R23 as 580mV but a quick calculation showed me that this is more or less the voltage, using Ohm's Law, expected at the junction of R22/R23, and it's a positive voltage NOT a negative voltage which would be the case if the detector diode was working because the diode anode connects to R22/R23 junction (not it's cathode as shown in the drawing above!).

When you're dealing with a faulty circuit things are rarely clear cut especially if you can't see the parts. If I was prepared to remove the hybrid and grind off its top all may be revealed, but in doing this there's a fair risk the hybrid would end up (more?) damaged so some experimentation is called for. Looking at the above hybrid circuit it would appear that RF current flowing through R15 is sensed by D3 which is forward biased via R22/R23 and R17/R15/R16 with C8 and C9 used merely to isolate any DC relating to D3. C8 and C9 also provide the possibility to use a DC connection to the diode making performance much less variable than it would have been using a DC blocking capacitor.

During the fault-finding exercise I'd discovered the last setting had probably been in the 80m amateur band. This poses an interesting question which is why this equipment had seemingly been dumped in a damp garage to moulder away. One answer is that during its last use an 80m transceiver had been accidentally switched to transmit. What followed might have been catastrophic, and no doubt had been accepted as such, when the meter displaying RF output had dropped to zero. Hybrid RF amplifiers are hard to source and if a replacement was found, it would cost a lot more than the price of a cheap modern signal generator.

What might have happened is not as bad as it could have been because, with normal RF output present, but uncontrollable via the AGC circuit, the damage seems to have been isolated to only the detector circuit. My guess is a high voltage passed through the attenuator and zapped the diode, D3. This failed short-circuit and immediately the 200 ohm resistor had gone open circuit, leaving R15 intact. The overload probably wasn't severe enough to kill the RF amplifier.. but all this remains to be verified. Read on.

 

 Above is my experimental detector circuit (in retrospect this was the Mk1). I'm using a small Schottky VHF/UHF diode, type 1N5711 connected via a 33pF Suflex capacitor and 200 ohm resistor to the RF output with ground return for the diode bias a 1Kohm resistor. A 120pF Suflex capacitor decouples RF present at the diode anode where it's soldered to the live end of R22. I switched on the equipment and instantly the meter read two-thirds scale, and the voltage at the test point (top right) read -5 volts (the manual tells me it should be -3 volts, but of course lots of pot twiddling has probably left the AGC loop miles out of adjustment and the parts I've used are probably not optimum).

Checking the meter reading as the attenuator was rotated gave me the following results with reference to the 0-10 meter scale:

Max attenuation up to 0dBm = 2, +10dBm = 3.4 and +20dBm = 6.8

Having worked out the likely key reason for this equipment to have been relegated to a damp garage, I need to consider if any other fault could have been introduced at the same time as the damage to the RF output hybrid. One consideration is the output attenuator. Whatever it's setting, 0dBm, +10dBm or whatever could have resulted in damage to that specific range. I measured the input impedance to ground and found it was basically 50 ohms from max attenuation to 0dBm, when it rose to 60 ohms. In the +10dBm and +20dBm settings it was infinite. As there's a 100nF capacitor linking the attenuator output to the RF output socket, any 50 ohm load there wouldn't show up.

I've not yet seen any of the lamps to the right of the meter illuminate so I'll need to check them. It seems these are all warning lamps but after checking their continuity I think all are OK and after some experimenting I noticed the top right (warning) lamp came on so the circuitry for these is probably OK.

Looking now at the three lamps (DS4, DS5 and DS6) on the left side of the meter. The lower lamp, DS6, is always on which seems wrong as these lamps are supposed to indicate the specific scale relating to the RF voltage which, since adding the external diode detector has varied sufficiently to swap over the lower/centre lamps (DS6/DS5). In fact the logic behind lamp selection is not complicated, being any odd numbered attenuator setting (= any voltage range starting with 3) should flip illumination from the lower (= any voltage range starting with 1) to the centre lamp (= any voltage range starting with 3). The control connection indicating odd/even passes from the mode switch A1A1 at the rear of the attenuator via pin 8 of A2 to a set of three transistors on the pcb. When the control line is high (= any voltage range stating with 1) the lower lamp (0-10) illuminates and when a ground is subtended from the mode switch to Pin 8 the lamp for the 0-3 scale should come on. The mode switch on the rear of the attenuator is coupled to the attenuator selection switch shaft and has three separate functions, one of which is to forward a ground connection whenever an odd range (= a voltage range starting with 3) is selected.

I'd looked at this mode switch previously when tracing another of its functions (to do with meter readings) and it looked OK, but having successfully traced the connection from A2 Pin 8 all the way the this mode switch, I realised the wiper responsible for carrying ground for odd attenuator settings was missing. There are two ways to get at the mode switch. The obvious one is to detach the attenuator, but as the knob refuses to be parted from the attenuator shaft I decided to try the most awkward which is to unscrew the A1A1 assembly from the back of the attenuator left in-situ. From the messy soldering at that blue potentiometer I'm not the first to investigate this area. As you can see below the end wiper was missing, and in fact one of the plastic pips securing the centre wiper had broken also. The plastic disk has two sets of pips so I moved the good wiper and the wobbly centre one to the spare set of pips. Finding a suitable new wiper could have been easier but I found that a contact from an edge connector provided something suitable so glued that in place. I then used an epoxy resin to complete the job.

 

 

 

 

 

 

 

Once I'd assembled the mode switch I found the new spring contact wasn't flexible enough and the pair of adjacent tracks was not being shorted by the new spring contact. The problem was that the ends are joined and either one side or the other made contact with the track but the contact needed to be able to twist slightly for the second track to be touched. I suppose I could have used a second piece of the material and reversed it with the join under the glue but I removed the spring and instead found a flexible relay contact of the right shape (below) and which is joined at the base allowing the two springs to work independently.

 

 
 

 

Left, the Mk2 repair: I searched through lots of relays until I found a miniature type having a low enough current capacity for the spring to be small and flexible enough for the job. I used one of the pair above which looked OK. Note, on the left there are two original wipers not four. Each has two prongs and pairs are joined at their base. Having twin prongs reduces the tension required to bring them into contact with the fixed plate carrying the tracking. I used one of the above but I needed to separate the two prongs slightly so these would make contact with adjacent printed circuit tracks. By allowing enough length in the spring the operating tension was lessened. The pencil mark tells me the centre line.

If anyone needs to carry out this repair, or a similar repair to other mode switches, you'll find a circlip mounted behind a tensioning spring and washer. The circlip has very small holes for using circlip pliers. The holes are so tiny I had to spend half an hour carefully filing down the ones you see below left. I eventually removed the pot to gain better access to the securing screws for the circuit board and access to the shaft of the switch. Note that the shaft of the pot has an open slot so you don't have to bend the mating spring joining the switch to the pot. The plain side of the spring slides through the hole in the centre shaft and the hooked side of the spring fits into the slot in the shaft of the pot. If you fit it the other way it will eventually slip out.

 

 

 

 

 

 One step forward and two back! I started to assemble the repaired mode switch and the adjacent plumbed-in coax feed to the oscillator came adrift. I'd noticed it was very wobbly when I first removed the bottom cover and the reason turned out that the coax bulkhead socket had sheared off. The coax complete with plug and socket came away from the back of the oscillator so I'll need to remove the whole oscillator assembly, remove parts of the rear enclosure and fit a new socket. Looking into the hole I could see the loose nut and a small loop to which the socket had been soldered. I suppose I would have needed to remove it anyway because I need to fix the tuning mechanism.
 
 

 There was a slight possibility I could access these eight screws and repair the broken socket without having to remove the whole assembly but the screws were too tight so not feasible...

At first sight I'd have to unsolder lots of wiring to remove the whole thing completely, but this wasn't necessary. A better solution would have been to pull out the riser to which the cabling is attached, but it seems one can sufficiently loosen the whole oscillator assembly without needing to remove it.

 

 

 

 

Below, it was possible to detach the second solid coax lead remove the four fixing screws, and rotate the whole assembly to access the side plate and gain access to the damaged output connection.

 

The oscillator turned out to be a very tight fit and I had to unscrew the two coax plugs at the RF driver module to gain a few more mm of wiggle room. No unsoldering because the cables allowed enough movement to access the side inspection panel. Once the oscillator was loose I upended the chassis so the inspection panel was horizontal, removed the eight long screws, lid and two RF gaskets, to expose the damage. The right-angled bulkhead socket fixing thread had broken off and subsequent movement had detached the RF connection to the circuit board allowing the nut and a star washer to fall into the box carrying the RF amplifier circuit board. I initially looked for a new socket but was unsuccessful. I noticed the remains of the thread had a flat (this is not used as the mounting hole is circular and provides room for the added wires). I used a couple of stiff wires from a CAT5 socket which I'd cut open looking for flexible gold-plated springs to earlier repair the mode switch. These are made of gold flashed brass and once soldered in place made the old connector a force-fit in its mounting hole. I added superglue to prevent the thing moving and soldered the ground wires to the circuit board ground plane, then soldered the RF output wire to the centre conductor (see the pictures below).

Then came the job of refitting the oscillator (I'd forgotten to fix the multi-turn mechanism.. but that can be done later because I'm not happy yet about what exactly is wrong with it). The various solid coax cables screwed back into place including the repaired cable which now feels nice and secure.

Slightly puzzling is that there are two sizes of RF connector nuts. Either this is an HP manufacturing arrangement or some solid coax leads have been replaced during the life of the signal generator.

 

 

 

 After refitting the repaired oscillator I continued with the repairs to the circuit board fitted to the rear of the attenuator. I'll describe these as progress unfurled... With the repaired mode switch and its pcb in place I added the potentiometer, having fitted three new connecting wires because these were pretty fragile and it would have been virtually impossible to re-connect these in-situ if one or more had broken off. I then re-connected the set of push-on wires from a sketch I'd made, which completed the overhaul, and tentatively turned on the equipment. Immediately I noticed a new lamp was lit and switching the attenuator backwards and forwards proved the repair to the scale indicator lamps actually worked.

My satisfaction at seeing the centre scale lamp illuminate for the first time was short-lived because the RF output meter showed zero RF output. Maybe I'd inadvertently pressed one of the switches, but the RF on/off switch said ON and none of the push buttons resulted in any meter movement, although I did notice a slight kick upwards with the attenuator turned to +10dBm. I couldn't believe the repaired mode switch area was bad so could it be a random failure? But no, I ruled this out as being too much of a coincidence.. more likely I'd disturbed an RF connection.. perhaps one of the rigid coax cables was now bad? The frequency display indicated a frequency in the 4-8MHz range, and twiddling the tuning knob altered the reading, so at least the oscillator was working and the repaired coax connection was fine.

I checked the RF output at the test point in the corner of the RF output pcb. Nothing.. this lines up with the meter reading so the metering circuit itself was probably not to blame. Time to use an oscilloscope and see exactly what was going on. I found there was RF at the output from the hybrid, but only around 180mV RMS instead of something like +20dBm which is around 2 volts. Could the rigid coax between the driver and the output stage be bad? The driver pcb output into the coax measured 20mV, so maybe a short circuit in the rigid coax? I also checked the auxiliary RF output and this was OK as was the RF output directly from the driver hybrid which measured about 500mV RMS. The problem was in the circuitry between the driver hybrid and the output into the filter box which seemed to be around 20mV RMS. Not thinking too clearly, and being interested in looking at the filter pcb for the first time, I opened up the screened box, detached the pcb above the filter assembly and made some resistance checks. The filter pcb was very interesting and I realised that all the slide switches and relays might have a bearing on the stuck 4-8MHz range I'm working with. But... I gave this area a clean bill of health in respect of zero RF output and put everything back.... but see later.

I decided to think logically and came to the conclusion that the reason for the lack of RF was connected with the modulator on the RF driver pcb which is built from lots of diodes. The control voltage to the modulator (which is described in the documentation as an attenuator) was probably the culprit. I reckoned the control voltage was attenuating the RF from 500mV to 20mV. Ordinarily the modulator, at least in CW mode, should be driven by the AGC circuitry to produce the correct RF level into the main attenuator.

I measured the control voltage and saw it was around 1.5 volts, but somewhere I'd read that circa 15 volts should be expected. Back to the schematics.. and I found the vernier control was the primary driver in producing the control voltage. Of course the vernier control was the potentiometer on the back of the main attenuator that I'd fitted not too long ago.

 

 

 Making measurements around the pcb at the back of the attenuator is tricky because some connections are hidden and pin numbering is slightly odd. Also, resistance measurements can be misleading because of various external circuit connections and it was at this point I found the vernier potentiometer had a dead spot at one extremity. I also discovered the pot was possibly wrongly wired (some time ago I noticed the wires at the pot were roughly soldered as if someone was messing with the circuit... perhaps looking for a fault?).

After 30 minutes of checking I found the pot was indeed wrongly wired and corrected this. I turned on the equipment, confident that all would be well.. but no.. the RF output although slightly greater (not quite as close to zero as it had been) was still miles too low, being down in the first few meter graduations. In fact.. the wrongly wired pot hadn't previously resulted in zero RF output.. but it hadn't actually altered the meter reading as it should have done... so that had been a fault I'd overlooked after restoring the RF meter readings.

Back to that A1A1 pcb.. and after comparing the circuit diagram with the wiring, and the pcb tracking, decided all was now well as everything matched. Could the problem be the wiper springs? I could clearly see the springs were in solid contact with the gold tracks and I'd already squirted switch cleaner on them just to make sure, so I convinced myself it must be an open circuit.. maybe corrosion had opened a gold flashed copper track... so reluctantly removed the whole assembly for the second time. Having got decent circlip pliers it was much easier this time, and I'd anyway decided to fit a new pot without a dead spot...

I carefully checked the pcb. It looked OK. I looked at the plastic disk with the three springy contacts and all looked fine. I could see nothing wrong with either the circuit board or the mode switch, so what on earth was the fault? I connected my multimeter with croc clips to the vernier tracks, and set to buzzer held the plastic disk against the tracks. Nothing heard.. so I jiggled it and examined the thing using my magnifying goggles. The spring contacts were solidly shorting the tracks but continuity was absent. Very odd indeed so maybe there was a break between the two prongs... surely not?

 I placed the multimeter against the prongs and found no continuity. I carefully placed the multimeter against the pair of springs on the same prong. Again no continuity. I looked at the prongs with my magnifying goggles.. perfectly shiny and undamaged, but then I realised what had happened... To secure the springs on the opposite side of the plastic disk I'd used a spot of superglue before using epoxy cement to finish the job. Superglue has a very small surface tension and flows across anything on which it drops. Gold flashed springs provide no resistance to flow and the glue had formed an incredibly thin insulation layer. I have a tube of superglue remover and applied a little to the springs then cleaned it off to be rewarded by continuity between the prongs and, after holding the disk against the track, I could see continuity across the vernier circuit.

I reassembled the pcb then found a decent 1Kohm linear pot and fitted this in place of the bad one (which not only had a dead spot, but whose resistance readings I later discovered were very intermittent). After a quick check that everything was connected correctly I switched on and RF output was restored and for the first time.. twiddling the vernier control made the output go up and down. Now.. what's the next problem?

 One of the trickiest jobs on the list is to overhaul the A9 assembly. This is a complicated thing that would have benefited from microprocessor control, but thankfully (for repairers) early microprocessors complete with complicated custom chips and (now long obsolete) special logic chips were still a gleam in the eyes of designers when the HP8640B was under development. I suspect so much cash had been invested in it that, although a digital design was possible, HP continued to manufacture the analogue version for ages. Below... A9 in-situ. The black markings are to aid reassembly later.

 

 A9 deals chiefly with switching the frequency range, and handling modulation... a very clever mechanical design which allows the switching of one set of circuits by two separate control knobs forms the basis of the assembly. In addition to gear wheels, no less than three mode switches are involved. Alas, the clever design is let down by deterioration of parts due to ageing. Firstly, in my example, the main gearwheels which have brass centres onto which are pressed plastic gears, are pretty fragile. As I'd spotted the damage I didn't attempt to force anything, so touch wood, I can repair rather than renew. As with most old equipment, certain parts are critical.. and the whole of A9 falls into this category.

Removal looks straightforward, but you do need to know in advance that it's quite possible, because there's not much wiggle room. One important matter is the link between A9 and the filter cam shaft.. the metal linkage needs to be in a vertical position for A9 to slip out (the coupling was very hard to turn). Once out the gears and mode switches can be examined properly. Clearly, the gearwheel plastic material has shrunk over the years, but this appears to be random because in my equipment shrinkage varies even between identical gears. I've heard it said that gear cracking is aggravated by expansion of the brass centre hubs and looking at a couple of cracked gears.. one is cracked at the hub and not in the gear itself and another has cracked at the hub with the crack extending to the teeth which implies the brass centre probably is involved in the start of the cracking.

 

 I tackled the two gears differently. In one case (the leftmost of the set of gears above), I ran superglue into the crack, into which I added small pieces of metal as a filler (see below). The reasoning was that the plastic is unlikely to expand and adding material would be better than relying solely on a large thickness of glue. The metal washer adds significant strength and will not interfere with operation.

 

 

 

 The worst damage was to one of the two bevel gears. One crack extended through to the teeth making it unusable because the gap must have been the best part of a mm in width at the teeth. A second crack wasn't too bad. The repair solution was to press out the brass centre bush after removing both grub screws. This was a risky business with the severe cracking but I initially left the gear overnight in the freezer hoping the brass would shrink a little and possibly relax its adherence to the plastic.

If anyone would like to copy the method I'll explain it here. I found a metal collar (this can be a large nut) whose inside diameter was slightly larger then the diameter of the brass bush, plus an M5 hex bolt whose circular head was just smaller in diameter than the diameter of the brass bush and whose length was short enough to allow the brass bush to press out. I used a small engineering vise as the press. The collar, gearwheel and bolt were positioned centrally and the vise very carefully tightened. Once the assembly was under pressure I waited and as the pressure slackened, gradually increased it to the previous amount. The end result was the brass bush dropped out leaving the plastic gear unscathed... in fact as soon as the brass bush was out the main crack began to slowly close up, and after 30 minutes the main crack was almost closed. I placed the shoulder of the bevel in the side of the vise and gently fully closed the crack whilst applying superglue. I also applied superglue to the second less obvious crack which had closed by itself.

Leaving the glue to set I measured the brass bush. This was knurled and the overall diameter measured 11.37mm. I also checked the size of the hole in the glued gear to find it was 10.77mm, some 0.6mm less than the bush. The next task was to reduce the diameter of the brass bush to something close to 10.8mm. I used a digital vernier tool to make the measurements but I suppose trial and error would be possible.

As the reduced brass bush needs to be as circular as possible and not having a lathe, I clamped my bench drill horizontally in a large bench vise. The bush was held on a bolt, having a diameter very close to that of the hole in the bush, with washers either side. Using a fine flat engineering file I reduced the size of the bush until it was about 10.8mm, not quite small enough to fit the gear. The inside of the hole in the gear was slightly rough because of the knurling on the bush so I carefully scraped the roughness with a scalpel (you could use a round file) until the bush fitted the hole. It's important not to apply force otherwise the superglue might fail. Once the bush has slid into the gear you need to double check the position of the grub screw holes which must line up perfectly with the tapped holes in the bush, then superglue the plastic to the bush.

At this point I was very pleased to have completed the repair... but alas, I hadn't appreciated one important point. To continue... because the brass bush was now smaller by over 0.5mm the threaded section of the gearwheel and the threaded hole in the bush were no longer continuous, neither was the position of one hole in the gear perfectly lined up with that in the bush. This next step is pretty critical. If you forced the "0.05 inch" grub screw into the brass bush it would almost certainly cross-thread and be likely to split the plastic or jam in the bush. The aim therefore is to remove the thread from the two holes in the plastic part of the gearwheel. If you try to do this without care the plastic might split. I used a sharp drill, turning it carefully by hand with the minimum stress because the repaired crack was centred on the grub screw hole in the gear (the weakest point).

Once the hole was slightly enlarged I used a mini drill fitted with a round burr tool to grind away the thread immediately adjacent to the brass bush. This is less risky than trying to remove the thread completely. By trial and error I was able to fit both grub screws backwards from the centre of the bush so that both were clear of the centre hole. Of course there's a risk that you'll forget and sometime in the future try and remove the grub screw completely and crack the plastic, but you could place a note to this effect inside the case?

 
     
     

  The first job is to remove the centre bush and reduce its diameter to almost fit the plastic gear

There are two of these gearwheels with the bevel but the other, which is not cracked, has no securing grub screws because it's designed to freewheel on its axle.

.

The second job is to clean up the centre of the plastic gear and re-fit the bush without cracking the repaired plastic.

Finally, clean up the thread in the plastic and fit the pair of grub screws (from the inside of the brass bush not as normal) ie. Insert the 0.05 inch Allen key through the hole in the plastic gear and attach the grub screw with tweezers inside the bush and unscrew it into the thread.

 I suspended A9 overhaul at this point so that glue can set, and delayed fixing the two damaged mode switches, but this should be easy because I already repaired the one at the rear of the attenuator...

The next major task on the list is to repair the AM switch. When the HP8640B was removed from the barn (ie it's a "barn find") I guess the owner tried the various knobs and switches. He'd mentioned he'd plugged it in and seen no display.... The front controls had suffered from damp and aluminium corrosion had seized many of them. Pressing the AM slide switch had broken off the lever which is quite small and, being plastic, had limited strength. Fortunately the broken lever was still present in the slot.

 

 The AM slide switch is not a discrete component. The top part carries three wipers, just like the ones in the mode switches that connect across gold flashed printed circuit track on one of the motherboards. Although the similar FM slide switch is riveted to the motherboard, the AM switch top is held in place by six nuts and bolts, enabling the top to be detached. Five of the nuts are readily accessible once the mains on/off switch and the modulation control plus the adjacent BNC socket have been removed, with the sixth a little bit awkward but possible. Slightly tricky was detaching the BNC connector because the securing nut was seized, also from aluminium corrosion, but by fitting a spanner to its rear flats I was able to jiggle it until freed.

Once the slide switch top was detached I cleaned it and glued the lever back in place. This however is only the beginning of the repair because superglue will not have sufficient strength to keep the lever attached. The key to this repair is to minimise the overall size of the lever because there's very little clearance in the way it's fitted.
 

Pausing for a moment... there's a couple of alternative solutions for a bad AM slide switch. One is to dispense with the original lever and glue an alternative metal one in its place. The second, which will need to be thought out in detail, is to procure a discrete slide switch having five positions and mount it as close to the motherboard as possible but this will mean either modifying the lever on the new switch or cutting the front panel. A warning here.. not all slide switches have the same switch configuration. The HP8640B needs an ON-ON-ON-ON-ON switch.

 

 

 Once the six securing nuts and bolts have been removed three main pieces lift off. The centre part carrying the spring contacts is shown in its assembled position. The two edge pieces are identical** and carry two slots. The wide slot mates with the centre piece and the narrow one is a guide for a springy wire with a raised pip (clipped to the front edge of the centre part) which locks into holes (the dark areas) to provide an ident for the five positions of the slide. The broken lever is shown before glueing back in place at the front edge of the centre part.

** They differ slightly because the fixing screws, which have different spacing, are countersunk so that the lever can pass over them, and the edge pieces can't be swapped around.

 My solution was to strengthen the existing lever. Because of the tiny allowable clearances I needed a material having the minimum thickness combined with strength. After some searching in my collection of "useful" stuff, I found several candidates each of which I measured with a digital vernier. By far the best was a sheet of thin steel having a mere 0.15mm thickness. Because it's so thin it was easy to cut with scissors and, using my small vise, I bent a small piece into shape. It looks complicated but it was quite easy to do. Once I was happy with the fit I glued the metal over the top and along the lever. Pictures below...

 

 

 

 

 The metal has a thickness of 0.15mm (about 5 thou inch).

From memory I think it's a scrap piece of mu-metal which is sold in very thin sheets and very easy to cut and shape.

A nice touch is the side of the lever carrying the marking is the visible side once the switch is reassembled and the repair is hardly visible.

 

 

 
 

 

 The A9 module connects to this cam assembly on the left via a coupler. It in turn mechanically links via six cams to the filter pcb, below.

I'd already noticed that an inordinate amount of force was needed to turn the operating shaft when positioning the linkage vertically to enable A9 to be extracted. I was now concerned that the coupler could be damaged if I didn't investigate and correct the stiffness in operation. Of course this stiffness could be quite normal, but a check on the Net suggested something had probably seized.

 If you look closely at the filter pcb you can see six slide switches operated by black plastic pegs more or less in a vertical line to the right of the metal casting. These pegs are part of the cam assembly mounted vertically above the filter pcb and should be free to move sideways, however the whole cam is seized. Note the tiny relays at the top of the pcb. These are used to select High-Low filters (see later).

 

 

 Left and right are the bearings at the end of the cam shaft. You can see a dark residue (left) which is dried grease. This had turned into a glue which had locked the bearing causing the whole thing to rotate in its mount between the base and lid of the box in which it fits.

Switch cleaner dissolved the grease which I need to remove and replace with fresh grease. A suggestion is to also reduce the pressure from the metal plate acting as an ident for the cams as this normally introduces too much force on the teeth of the mating gear. Note the hairline crack close to the cam ident gear.

 

 After applying fresh grease to the bearings and the cams and slackening the two screws holding the ident plate I added a little superglue to the latter screws to prevent them dropping out. After refitting the lid which holds the camshaft in place the shaft turned smoothly so I returned to refurbishing the A9 module. The main task here is to complete work on the three mode switches. One has two wipers, the second three wipers and the one on the opposite side has four wipers. The first couple had one wiper each still fastened to the transparent disks and the other had all four still present. I fashioned new wipers and stuck these in place and decided then to remove the third to check its condition. I found that all four wipers were variously detached so I added a spot of superglue to each. The paper was to ensure the wipers were lined up as the glue set and to absorb any excess. The wipers were made from relay parts as before.

 

 

 Once the mode switch disks had been repaired I reassembled the A9 unit. Of course you need to ensure the correct switch positions are selected and each gear is secured primarily on the flat of its shaft so that misalignment can't occur. With both main shafts fully anti-clockwise the pair of mode switch wipers should be vertical looked at from the underside with the rear mode switch at 180 degrees to these. The pair of bevel gears should mesh closely and the spur gear should be in line with the other three.

The action of the main shafts now feels nice and positive, but I need to fit the repaired module, taking care to line it up with the cam shaft to see if it works correctly. Oops.. see the results below.. could be better?

 

 Range

 Low

High

 Low

 High

 Comment

 A8A3 logic

 512

 460

 1100

 4.584

11.015

 Decimal point?

 position 3

256

 230

550

 2.292

5.5

 Decimal point?

 position 3

128

 115

275

 4585

11017

 ??

 position 3

64

 57.5

137

 5732

13772

 Decimal point?

 position 4

32

 28.8

68.7

 2865

68681

 Decimal point?

 position 4

16

 14.4

34.3

 1432

03441

 Decimal point?

 position 4

8

 7.20

17.1

 7.159

17.208

 OK

 position 5

4

 3.60

8.59

 3.570

8.607

 OK

 position 5

2

 1.80

4.29

 1.790

4.303

 OK

 position 5

1

 0.90

2.14

 0.895

2.153

 OK

 position 5

 0.5

 0.45

1.07

 4.476

10.759

 Decimal point?

 position 6

 Without a decimal point the display was a bit mystifying but having noted the "overlap" range frequencies given in the user manual, things became a lot clearer. It seems the HP8640B is as complicated in its logic circuits as in its mechanical bits and pieces. The Counter Time Base Assembly, A8A3 is responsible for handling the decimal point position, which it does by reading the range setting (=band code), a set of data inputs A-E (Counter Mode). Complications arise when the push buttons for giving additional accuracy are pressed (ie x10 and x100). For each range only a single data input (A to E) can be at logic LOW. Assume the x10 and x100 buttons give the correct output, both are HIGH. The calculation on the position of the decimal point isn't a static logic level because a new calculation is performed at a regular rate set by the Time Base. I removed the upper board of the two mounted in the box at the right of the display. This is the one carrying the display, A8A3, and underneath is A8A2 which carries the decimal point logic. Noteworthy is the fact that some of the i/c's have rust on their legs, but otherwise, being gold flashed, the printed circuit looks OK. Measurement of voltages around the legs of i/c's is not straightforward when things are powered up because much of the circuit is either driven by a "clocking" source, or perhaps asynchronous signals, rather than being static voltages. The former will be averaged by a DC voltmeter and therefore pretty meaningless when it comes to checking logic states.

I'm looking for unusual static voltages. Basic 74 series logic, on which the HP8640B, mainly relies, generally has two possible DC levels, being 0 or 1 logic states. These levels are roughly 0.5 volt and 3.5 volts, give or take, and once versed in circuit checking one can readily see an error due to a faulty chip. Whilst not a clear cut way of detecting a problem, it does point you to further analysis. Below, I've shown the decimal point decoder on pcb A8A3 and one of the display chips on pcb A8A2.
   
 

 Tracing back from the decimal point pin on each of the seven display i/c's I found that printed circuit track is routed to Pins 11 to 15 on the riser carrying the display board (eg Pin 12 for DP4 of Display U4 above). These are connected to U3 (Pin 14 for DP4 above), an SN74LS96N which is a rather uncommon i/c described as a "5-bit serial-parallel, parallel-serial storage register". Checking U3 dynamically is tricky but, because the chip has four control inputs handling its five identical "totem-pole" output circuits, and its outputs connect either to five identical displays, or with the display board unplugged, free from other connections, all outputs should test identically when unpowered. Using a multimeter set to diode or ohms I would expect to see each output to be roughly the same.. for example if one showed a forward-biased diode so should the remainder. In fact I found anomalies both in resistance and with a diode test.

I found a supplier offering a new i/c for 99 pence, so rather than continue testing with a major unknown, I'll just swap the 74LS96 and if this doesn't cure the problem, I'll continue testing. I have an i/c tester so before the new chip arrives I can remove and test the old one. This is not an easy job however. The lower pcb in the housing is held in place by five soldered connections to feed-through capacitors plus three rigid coax connections. The coax leads must all be disconnected before the casing can be removed. Connection to the adjacent motherboard is via a loose riser held in place by pressure once the casing is screwed back. Once the casing was free I cut the five feed-through leads and the board lifted out. I removed U3 and tested it.

 

 

 Left, A8A3 pcb less U3 at top right corner, removed from its case (above) with a view of the main chassis (below left) showing the three rigid coax leads (the right lead just pulls off whilst the others unscrew) and, right, the loose riser.
 

 

 

 Static tests showed the outputs were faulty and these only connect to the display so other i/c's on the board should not have been affected by this failure.

The chip is really complex so should I rely 100% on the test meter?
 .
 

 

 I believe this item is Option 001.
 

 Whilst I'm waiting for the new 74LS96 chip I'll tackle any remaining problems, one of which I spotted only when refixing some of the control knobs. This is another split gearwheel hidden away on a shaft protruding from this Modulation Frequency Oscillator box, however.. when I looked at this it seems quite difficult to get at and when I rotated the control knob to which it's attached the crack was equally spaced between the limits of operation and never actually engaged, remaining clear of the driving gear so I just left it in place.

 

 Finally, the new 74LS96N arrived.. I say "new" but the date code told me it was made in Week 25 of 1982. Anyway popping it into the tester proved it was fine but, because a week had taken place, I found it tricky getting all the parts back together. Eventually I'd used up all the screws and it looked OK, at least mechanically and I'd taken the trouble to remove dust from the red display filter. I switched on and the decimal point was....... OK.

 

 
 
 A little difficulty then ensued because, once I'd hooked up my scope to the RF output I discovered the RF didn't match the counter display, but after correctly re-aligning the filter cam to the range switch all was fine... except the top three ranges displayed incorrectly. The RF output was OK, because I could see a 128MHz trace on my 100MHz scope, but the counter display thought it was only 1.000MHz. So I checked further...

 

 

 Range

 Tuning MHz

 Comment

 RF Output Volts RMS

 512

 4.000

 wrong display

 NA

256

 2.000

 wrong display

 NA

128

 1.000

 wrong display

 9.5

64

 64.000

 OK

 7.5

32

 32.000

 OK

 8.4

16

 16.000

 OK

 11.1

8

 8.000

 OK

 12.0

4

 4.000

 OK

 15.6

2

 2.000

 OK

 15.8

1

 1.000

 OK

 14.0

 0.5

 0.500

 OK

 13.4

 The GDS1102 scope runs out of steam at something over 128MHz so I've put NA in the RF output column.

Once everything gets sorted out I'll use my DSA815TG to make accurate tests.

Clearly, there's a problem, though not with the signal generator.. but maybe the frequency counter?

 Reading the repair manual for the HP6840B I spotted a truth table relating to the logic for selecting the right set of dividers for the counter. This suggested that the reason for the bad display was Pin 9 of U6 on A8A3 was stuck at logic one. This pin is tied via 10Kohm to the 5.2V rail but is grounded via S2 on A9A1 in the top three ranges. I traced the signal where it exits from the track on plug P2 at the top of the front range switch pcb via a flat cable to the local motherboard. It then passes via a loose riser to a 1mH choke on A8A3 to Pin 9 of U6. Initially I found the path to be open circuit, but after re-plugging P2 into the motherboard and re-checking, the path measured OK, so I turned on the power and was pleasantly surprised to see 512/256/128MHz all present on the display. For anyone interested I've reproduced the relevant section of the truth table below.

The fault is that 128, 256 and 512MHz are displaying as 1, 2 and 4MHz respectively and from the table you can see that this can occur if U6C9 sticks at logic high. The codes for the higher frequency ranges are then the same as those for 1, 2 and 4MHz. This can be due to one of a number of different reasons.. an open circuit, a faulty mode switch (S2 on A9A1), a faulty U6 i/c or even a bad ground connection. The various i/c's U6, U7 and U11, in addition to U3 and U5, are used to drive the dividers which develop the appropriate RF frequency from the main oscillator to drive the counter. If there's a fault within this area you can get a true RF output but a false display (which was exactly the case here).

 

 RANGE

 U6B4

U6B5

U7B5

U11B4

U6C9

U11A2

U6C10

U6B6

U11B6

 1MHz

 H

H

L

H

H

H

H

L

H

 128MHz

 H

H

L

H

L

H

H

L

H

 2MHz

 H

H

H

L

H

H

H

L

L

 256MHz

 H

H

H

L

L

H

H

L

L

 4MHz

 H

H

H

H

H

H

H

L

H
 512MHz

 H

H

H

H

L

H

H

L

H

 The next task is to measure the RF output and see if it matches the attenuator figures and to see if it maintains the correct amplitude across the whole frequency range. I might see if I can improve the range switch before I start. Mine is in a dreadful condition because it has been forced and the lever has broken off. Apparently the knob design varies because I found a picture of a pristine example.. below. The range knob is completely different to mine which used to be a similar shape to the peak deviation knob to the left. Can you spot something odd, presumably overlooked by the last owner of this model? click to see the answer

 

 

 Here's before and after pictures... so far.

 

 
 

 A new knob to replace the broken one.

 HP designed the knobs to include a thin layer of aluminium covering the plastic surface before adding a stencilled coloured background with black printing. Maybe this was done to improve the life of the knob but alas, damp results in corrosion to the aluminium and the oxide destroys the finish. The modulation frequency knob doesn't use the same finish and has outlasted the others.

You can also see that damp attacks the frame of the equipment resulting in a very scruffy appearance. I used emery paper on the frame and cleaned up the front. I found the best method of finishing the cleaning of the front panel was to use a rubber on the end of a pencil which nicely removed blemishes. I need to slightly increase the font size and use bold to make the numbers more legible. The method I used for the range dial was to draw a circle (PhotoShop), then type each number, rotate by 30 degrees and type the next. I used a definition of 600dpi and had to keep cropping the drawing as it increased in size every time I flattened the result between numbering. Printing was straightforward. I measured the size of the knob surround, noted the auto-print size and then applied a percentage reduction to match the print to the size of the knob. A thin black surround plus a deepening of the colour to match the other knobs might be an idea for the Mk2 version...
 I was finally at the stage where I can test the 8640B using a spectrum analyser. I did the tests with the lid off the RF compartment, but using a level of 0dBm or thereabouts this doesn't matter too much. I checked each range in turn from low to high and found a couple of interesting facts. Firstly the three lowest ranges were poor in terms of harmonics (I'd previously noted the sinewaves for the lowest ranges were distorted somewhat) and the remainder pretty good, although switching from 256MHz to 512MHz failed to make any difference I now realise that it won't make any difference because I'd need a doubler.. Option 002). Amplitude-wise the signals were in the same ballpark, although I haven't yet attempted to set up or check this aspect.

 

 

 All the measurements were much the same. This shows a 256MHz CW signal over a spread of 200-300MHz but extending the scan to 1GHz showed no discernable harmonics.

 

Below is a picture of the front of the 8640B for the output shown here.

The output meter indicates 0dBm which doesn't line up with the picture.

 

 Below, harmonics were only obvious in ranges up to and including the 2Mhz range. Is this a fault? Well, I looked at the circuit diagrams of the filter module and the four filters for the lowest ranges all looked similar, and each is followed by a second (common) filter to handle high numbered harmonics. I then looked at the specification which made bold statements regarding spurii, particularly sub-harmonics and non-harmonically related signals. Of course, because the 8640B doesn't use mixing processes it won't be subject to these. Instead it relies on a single high stability UHF oscillator and to obtain suitable output frequency ranges, the oscillator is divided by ecl (high speed.. emitter coupled logic) circuitry whose outputs are square waves. Square waves are rich in harmonics so will need converting back to sinewaves. This is achieved through a set of filters, so the harmonic performance will be directly governed by these filters. HP specify a harmonic output of only better than -30dB at 1 volt (RMS). This latter voltage equates in a 50 ohm system to +13dBm so -30dB represents -17dBm which doesn't sound great.

The pictures below are not ideal because the metal cover over the RF amplifier is not fitted, neither is the outer case... and the spectrum analyser bandwidth is set quite high at 30KHz and more. The 2nd harmonic at 500KHz looks like it's 25dB down and the 3rd is 36dB down. The 1-2MHz range is better as both 2nd and 3rd harmonics are almost 31dB down.. not brilliant but just within spec. The good news is harmonics are very much lower in ranges greater than 4MHz. Thinking about filter design... the designers provided a decent overlap between ranges so it's actually not just very difficult to design a filter having a good performance within a specific range, it's impossible to design such a filter. For example the filter for the Range covering 16 to 32MHz cannot remove the 2nd harmonic of 16MHz at the low end of the range without removing the desired 32MHz element when tuning to the high end. To get around this each range carries two filters which are selected by relays driven by circuitry governed by a pots coupled to the main oscillator tuning gears. In this way either a low band or a high band (and even both) is appropriately switched into circuit. If you listen carefully when tuning you can hear a faint click from the relays at roughly centre frequency. This isn't the case though for the ranges up to 8MHz where HP designers reckoned the 2nd harmonic would be at a sufficiently low level to not bother with the added complication. In essence, they reckoned the sinewave shape up to 8MHz would be good enough to result in the 2nd harmonic being better than 30dB down.

 

 

RF Output, Range 0.5 to 1MHz at 500KHz

Oscillator tuned to 256MHz & divided by 512

 RF Output, Range 1 to 2MHZ, at 1MHz

Oscillator tuned to 256MHz & divided by 256

 RF Output, Range 2 to 4MHZ, at 2MHz

Oscillator tuned to 256MHz & divided by 128

 Reading the Hewlett Packard documentation regarding the performance, but between the lines, an interesting comment is that after changing the modulator or its pre-amp you should check for harmonics "particularly at 4MHz in the 4-8MHz range". The indication may be that you shouldn't be upset to see harmonics below this range... which is exactly the findings above. Fortunately the 2nd harmonic in the higher ranges was not really noticeable, but if it had been a problem the HP solution is to swap resistors to reduce signal level in the divider chain. This applies also to the lower ranges, but I'm not going to bother about this...

The next stage in fault-finding is to look at the loop carrying the meter amplifiers with regard to the RF levels. Hopefully this will explain why the 256MHz signal set at 0dBm has a level of over 6dBm, above.

I'll just go over the way the ALC (automatic level control) loop operates. The output amplifier, driven by the modulator (these are the gold hybrids in the RF compartment) puts a well defined RF level into the attenuator. This level is +10dBm so that, for example, when the attenuator is in the 0dBm setting and the vernier control is adjusted to make the RF level meter read exactly 0dBm, you'll get 0dBm RF output. That is excepting when an attenuator range of +20dBm is selected. In that case a (different) resistor is switched into the AGC loop that pushes the RF amplifier output up to +19dBm giving an extra boost of almost 9dB to the RF output over that achieved from the +10dB setting with 0dBm on the meter.

 

 The ALC loop is controlled by two voltages, Vdet and Vref, which together control the modulator (essentially a voltage controlled attenuator) to form a steady state closed circuit. Twiddling the Output Level Vernier or switching the extra 10dB control (actually 9dBm), or indeed turning RF off, will manually change the steady state. Changes in RF output due to varying conditions (resulting for example from small changes in RF levels whilst tuning) will automatically maintain the steady state. Preset controls will determine the relationship between the manual or automatic loop changes and the loop steady state.Click here or the schematic above to read the PDF for power/meter adjustments.

 After thinking about the possible reasons for the slightly high output, I decided to first tackle the non-linearity. That is the poor frequency response of the home-brew detector. The one embedded in the hybrid uses DC coupling to the diode detector so will be much less frequency sensitive than my external detector which is coupled to the output via a 33pF capacitor. Below I've shown the effective resistance of the 33pF capacitor compared with one of 15nF. The diode is in series with a 200 ohm resistor which has the effect of reducing the effect of the capacitor perhaps up to a few hundred ohms, but once the frequency has dropped to 4MHz the capacitor has a significant effect. The rectified voltage will be less so the loop will then bump up the RF level into the attenuator. Choosing a 15nF capacitor shows that the 200 ohm resistor significantly damps out the variation. Inside the hybrid a 10nF coupling capacitor is used to block the diode DC bias so 15nF should be a good choice for the external detector (this was the Mk2 version).

 

 Freq MHz

 500

 250

125

64

32

16

8

4

2

1

 0.5

 MHz

 33pF

 9.6

19

38.6

73

151

301

603

1206

2411

4823

9646

 Ohms

 15nF

 0.02

0.04

0.08

0.17

0.3

0.6

1.3

2.6

5.5

11

21

 Ohms

With the new coupling capacitor in place you can see a significant improvement in the RF output level. The highest test frequency is nearly 4dB in error and is almost certainly due to stray inductance effects which increase the detector resistance, reducing the voltage fed into the AGC loop and bumping up the power slightly. A chip capacitor in place of the wire-ended type should improve this. Also notable from the reduction in output level is a reduction in the level of the second harmonic output. I tried a second experiment and the output level could be brought closer to 0dBm by slightly changing the 200 ohm series resistor.

 

 Freq MHz

 550

 256

 128

 64

32 

 16

 8

 4

 2

 1

 0.5

 O/P Level dBm

 -2.0

 +1.75

 +1.88

 +1.77

+1.72

+1.75

+1.62

+1.74

+1.8

+1.8

+1.86

 Meter dBm

 0

0

0

0

0

0

0

0

0

0

0

 2nd Harmonic

 -48

 -35

 -38

 -42

 -46

 -47

 -44

 -41

 -47

 -37

 -38

 2nd H attn.

  46

  37

40

44

48

  49

  46

  43

  49

  39

  40

 I tried modifying my home-brew detector to see if it could be improved. I used two 10nF chip capacitors in parallel in place of the 15nF discrete capacitor to reduce inductive reactance from its leads at 500MHz, and experimented with the series resistor, settling on a 30 ohm chip because that value fitted exactly between the new chip capacitor and the diode detector. Initially I tried two 100 ohm chips which brought down the level to 1.5dB then having put 30 ohms in place the difference across the whole RF output range was less than 1dB, reading typically +0.8dBm with the attenuator at 0dB and the vernier adjusted to a meter value of zero dBm. Because there must be some RF leakage I expect the discrepancy to fall slightly once the lids are back in place. I could have tried an sm schottky diode in place of the 1N5711 but the end result was good enough already.

 
 

 The modified detector circuit (Mk2) aimed at better frequency linearity and slightly higher DC output to get the AGC loop output voltage to drive the modulator hybrid such that the RF amplifier produces exactly +10dBm at all frequencies from 450KHz to 550MHz.

The AGC loop sets the RF output to +10dBm for all attenuator settings other than +20dBm, where the ALC circuit is modified by changing resistors on the A1A1 pcb (bolted to the rear of the attenuator).

For the maximum output of around +19dBm, a 10K resistor, R2 is bypassed and R3 of 28.7K replaces a 1K resistor R4.

 

Below a view of the underside of the 8640B (less the RF compartment cover)

 

 

I was determined to get the RF detector working as well as the original so embarked on the Mk3 design. I'd managed to reduce the error between the meter indication of 0dBm and RF output to within 1.8 to 2dB and decided to obtain a more sensitive detector diode. At +10dBm the RF voltage is about 700mV and as the highest frequency is 550MHz the diode (CR1 below)must have a very low self capacitance and a very low forward voltage drop (circa say 140mV). The bias voltage is set by R22 and R23 from the 20 volt supply (see below). This works out at about 613mV. The diode bias resistor is 1Kohm and the swamping resistor which averages the detected voltage across the whole frequency range is 30 ohms so the DC component of the diode current is about 470uA. Loss through the swamping resistor is negligible at about 14uV. The peak RF voltage is 990mV so there isn't a lot to play with to drive the AGC loop.

A detector diode having a capacitance of say 3pF has an impedance at 550MHz of about 100 ohms so in terms of a 50 ohm system will be pretty leaky. I chose a new diode having a capacitance of only 0.3pF (=1000 ohms at 550MHz). The old diode is a 1N5711 whose 2pF works out at about 150 ohms so the new one is much better. Once in place it's pretty obvious from its tiny size and fragility you can't transmit into it without ruining the diode. I wonder how many hybrids have been written off when only the detector is damaged, because the loss of the diode=immediate loss of meter reading which might have been assumed to be loss of RF output?
 

 

 

 Left: the diode uses an SC79 package (body 1.3mm x 0.8mm x 0.6mm thick with 0.15mm legs).It may be OK for a machine to stick it onto a circuit board but not easy for me to solder in place as shown below! I used the giant 30 ohm resistor to give mechanical stability.
 
 

 The perfect diode for the job isn't exactly the easiest diode to fit in place. The size of the body (SC79) in inches is 28 thou x 48 thou and the legs 6 thou each. You can only just see the cathode band and soldering it needs a steady hand. I used a thin silver plated wire to connect the anode to the bias point and hopefully this will absorb any mechanical shock. I wired it in place and switched on... the meter read zero. Well, I wasn't too surprised because seeing the cathode band was next to impossible even with my magnifying goggles. After unsoldering it no less than twice and failing twice to get the circuit to work I wondered if the cathode band was printed on the wrong end... so soldered it in place for the fourth time.. and success, the meter read 0dBm..... Below is the Mk3 detector.

 

 I checked the RF output at 5MHz. With the attenuator set at 0dBm and the vernier producing 0dBm on the meter, the spectrum analyser read 0dBm. Switching up and down proved the same power output within less than +/-0.5dB was achieved all the way from 450KHz to 550MHz.

I screwed the RF module lid back in place and looked again at the spec for the SMS7630-079LF. In small print it said "Marking:anode".

 I decided to check the various controls now that the basic RF area is working. Much to my surprise everything worked. I monitored output on my Icom R7000 tuned to 120MHz. First, a nice clean carrier, then switching the 8640B to AM I checked the two fixed modulation frequencies followed by the variable controls and all were OK. Next FM and this worked on all the relevant settings. I then switched the receiver to SSB and checked the carrier. This was nice and steady and was fine with both the Lock and the Half Digit switches doing as they should. The meter produced lots of readings in all three selector switch settings. I recorded lots of pictures, even checking the second harmonic of 500MHz at 1GHz.

Time to put the top and bottom covers in place and fit in my test gear racks for use on my next project.

 See a selection of RF output pictures

 Answer: The Range switch should show 16-32 not 8-16 because the knob is fitted wrongly.

 Return to Reception