Noise Source Mk2

 Having built a prototype noise source and weighed up its performance I'm almost ready to design a new version. This should enable further experimentation possibly including adding a third stage of amplification and maybe a high impedance buffer between the zener diode noise generator and the input to the amplifier. The first step will be to construct the two-stage amplifier. This will be done using a couple of printed circuit adaptor boards supported over a small piece of tin using 100nF decoupling capacitors.

If the high impedance buffer is a dual gate BF998 perhaps g2 can be used as an AVC point set to keep the output at a safe level for using with a Rigol DSA815 having a +20dBm max rating.. Although the output at any given frequency is about -40dBm, the total output power is greater than 0dBm, so if 20dB more gain is used the output gets a little high. I've already noticed the Rigol complains when the noise source is turned on. First though, I need to test the result of adding a FET buffer stage.



 The SOT143b FET is mounted on the tiny pcb, bottom left. It's supported by soldering the source, g1 and g2 resistors directly to the tin plate with the top right connection linked to the 100nF capactor feeding A1. The tiny zener diode noise generator is fed by the 1.3K resistor and connects via a 100nF capacitor to the lower right connection, g1. As it stands the green drain wire connection (top left) has not yet been added and g2 (bottom left) is grounded via a 47K resistor. Later I added a 100K resistor to the 7.2 volt supply to bias g2 slightly higher so the FET would draw more current.

 I measured the supply at 7.57 volts and the source at 280mV which means the FET is drawing 2.8mA. These FET devices are not dissimilar to thermionic valves and in this example is quite similar to a tetrode because as g2 voltage is raised the transistor draws more current. The benefit of using a FET is the device draws very low grid circuit current thus the zener noise output isn't damped as previously. As shown opposite however the circuit, being equivalent to a cathode-follower provides no voltage gain.

Initial tests showed a tendency to instability, probably due to the rough layout, but the HF response seemed much better as the higher frequency components of the noise from the zener are not being damped as much.

A quick check using an SDR set to a baseline of -155dBm showed a noise level of -116dBm from about 100KHz to a couple of GHz. However, my HP power meter showed exactly 1mW or 0dBm total power which is marginally less than before (1.15mW) without the FET buffer. In the test I measured the source voltage as 280mV and G2 voltage as 2.38 volts. These figures line up well with those in Fig 6 of the FET spec, supporting a drain current of 2.8mA.

It's possible to add some voltage amplification (say 10dB) by inserting a drain load resistor of say 330 ohms and using the drain rather than the source (with drain voltage about 6 volts).


 Testing revealed that the lid of the diecast box needed to be fastened surely to eliminate instability and at frequencies in the GHz region there was a some oscillation taking place, no doubt due to poorly laid out grounding. Looking at the results using the SDR.. the noise level is now uniform across the whole test range (which perhaps implies undesired resonance previously between the zener and the first amplifier stage), but less in amplitude (previously I'd measured about -90dBm noise and now -116dBm). The follower circuit does not in fact produce a unity voltage gain, but a small loss (confirmed by the total power reading of 1mW compared with 1.15mW). Providing the circuit is stable, arranging the FET as a low gain voltage amplifier might compensate for this.. Interestingly the zener power output seems to be relatively unchanged, staying much the same no matter whether connected directly to the first amplifier or the FET buffer.


 For the next experiment I connected the FET to provide a little voltage gain. The drain voltage measured 6.28 volts so with the 390 ohm load this represents a drain current of about 3mA. I set the SDR baseline to-160dBm and found the noise output was a constant -125dBm from 100KHz to 60MHz Then above 60MHz it increased to -100dBm where it remained up to 250MHz before dropping continuously with frequency. The circuitry of the FET stage must have an inherent resonance which in this layout makes it unusable.

The end result, after looking at the performance of the buffer stage in both configurations, I believe it's addition, at least as an amplifier is the wrong approach and reverting to the follower circuit and/or different changes to the original design may be more rewarding. What is clear though is the way the noise source works in practice provides a built-in way of monitoring its performance, particularly in identifying circuit resonances.


 I investigated the noise source using the circuit above with my SDR and found its behaviour to be fine up to 250MHz, but for an odd reason the noise output dropped sharply before re-appearing and fairly constant up to well over 1.5GHz, but at all frequencies the output level seemed to be weaker than the original (bufferless) prototype. I changed the buffer circuit, making it a follower rather than an amplifier and checked its operation again. During this change I removed the tin shield above A1 in order to resolder the input capacitor. I also fitted the g2 100nF decoupling capacitor directly to ground instead of to the A1 tin shield which I'd used for convenience. I measured its RF output and found it was much lower than before (bufferless). It had been exactly 1mW but was now reading only about 0.5mW. I removed the lid and was able to ground g2. On doing this the output rose to about 1.5mW so I cut the feed to the 100K bias resistor and found the output was now just a little under 1mW. A further test using the SDR showed the performance was virtually the same as before (with the FET as an amplifier) except the noise output was slightly greater over the whole frequency range but still exhibited the steep cut off at a little over 250MHz. The next test may be to directly ground g2. If that fails to improve the performance especially in terms of the apparent resonance at 250MHz I'll remove the FET buffer. I had assumed the zener noise may be damped by the input of A1 but experiments appear to indicate this may not be the case. Maybe the zener impedance is pretty low.. theoretically it might merely be simply related to its terminal voltage of 5.1 volts and the zener current of 1.6mA = 3K, although AC-wise the 1.3K load resistor must also be considered, making about 900 ohms. If the zener current was increased, what would its impedance be? Reducing the load to half would result in a zener current of 3.2mA and a zener impedance of 1.5K and a rough AC impedance of 450 ohms. These figures do not take account however of the RF impedance of the zener itself. As the designers didn't intend the device to be used as a noise source we can't directly use the spec for the zener, however there are some clues given such as its capacitance which is very roughly 100pF. Given that value we could say that this results in an RF impedance of say 15K at 100KHz, 160 ohms at 10MHz and only 8 ohms at 250 MHz. What isn't readily available however is any clue to the the level of avalanche noise from the zener at any given frequency. This might rise dramatically at VHF which might be the case given the initial results from the noise source? We can say though that the RF impedance of the zener is pretty variable across the test range I'm looking at. In fact, because of lots of unknown facts the only way to proceed is by experiment and so far I reckon the original circuit without the FET buffer is giving the better results.

One clue to the zener noise output is the fact that even slight parasitic capacitance at the zener-resistor junction does result in a significant reduction in noise output. That might mean that the zener capacitance of 100pF is not a relevant figure in its avalanche noise output. That being the case some experimentation with zener current may be a good approach. As I've already explained the total noise output from the noise source must not exceed a figure of +20dBm and to be safe let's say +10dBm. It's already 1mW or 0dBm and I've seen 1.5mW (+2dBm) during experiments so a figure of less than 10mW should therefore be the target. If -40dBm is the average level of indicated noise at any given frequency then -30dBm will be the final aim.

 After grounding g2 I decided the FET buffer wasn't providing any advantage at all compared with the original design so unsoldered it, and moved the zener closer to A1 input. I then checked the RF output and found it measured 0.9mW, a little less than earlier results but quite possibly because the battery voltage was slightly lower than it had been for initial tests. The table shows the noise at spot frequencies using my SDR-Play at the same gain settings throughout with the Difference figures added to compensate for changing SDR performance. No real resonance effects were apparent and figures are close, but slightly down on the first prototype, although as I suggested the battery voltage will be down slightly. The next step is to measure the true noise output using the Rigol spectrum analyser as the table merely gives signal strengths with the SDR gain settings fixed to give comparative readings. The noise source matches the SDR input nicely as inserting a 20dB attenuator reduced the noise level from-100dBm to -120dBm at 10MHz.











 Baseline dBm










 Indicated Power dBm










 Difference dB










  A quick check on the Rigol showed a reasonably flat noise output of -62dBm +/- 3dB from low frequencies up to over 1GHz with a very sharp resonance at 1.17GHz then flattening back up to the maximum Rigol frequency of 1.5GHz. Note that the indicated power level varied from -40dBm to -60dBm as the Rigol scan reduced from 3KHz to 300Hz. Later I retested using the Rigol and found the output showed a large spike of around 0dBm at 1.167GHz with the noise output across the lower frequencies (below 1GHz) of about 10 to 15dB above baseline. Clearly the noise source was working but the power output was being hogged by the 1.167GHz oscillation. By trial and error I found that decoupling the voltage at the output of the 180 ohm feed to the amplifier chips to two points of the chassis with 220nF capacitors cured the feedback and killed the oscillation. Immediately the broadband noise output increased to the level I'd seen earlier in the experiments of 40dB above baseline. Results are shown in the table below.


 Frequency span







 Baseline at centre dBm

 -84 @ 100KHz

 -87 @5MHz

 -81 @15MHz

 -80 @ 150MHz

 -74 @ 500MHz

 -74 @1.5GHz

 Noise at centre dBm

 -64 @100KHz

 -50 @ 5MHz

 -42 @ 15MHz

 -42 @150MHz

 -44 @ 500MHz

 -60 @1.5GHz








 The final results look encouraging enough to proceed on a circuit board prototype. A small imrovement to LF performance would result by increasing the three coupling capacitors from 100nF to 220nF, otherwise it looks pretty good. I also made some voltage measurements as shown below. From these readings you can see the total current consumed is made up from three components (1) two amplifiers 24.05mA (2) noise zener 1.92mA (3) LED 1.15mA which add up to 27mA. A1 output is drawing 10.7mA and A2 output 4.7mA, although the voltmeter may affect these measurement at A1/A2 outputs. Z2 wasn't fitted.


Having got the new noise source working reliably I decided to test a low pass filter using an SDRPlay but the results were very puzzling. The problem is that the SDR has lots of settings which need to be accurately matched to (a) the filter pass band and cut off and (b) the amount of noise provided by the noise source and (c) to give a meaningful picture of what's happening. Below is a picture of the set-up without noise then with the noise turned on. I was careful to adjust the various settings to minimise confusing software artifacts.


 Now an explanation of what is shown above & below. The SDR is giving a reasonably flat response of -139dBm +/- 0.5dB.from 31.5MHz to 39MHz and with no noise the curve in the first picture is produced with RF Gain minimum, Visual Gain -30dB and IF gain -30dB Manual. Without changing these settings I turned on the noise source to produce the second picture. There's a peak response at 31.48MHz of -95dBm before the filter begins to cut off with the 3dB point at about 31.8MHz. The baseline after the peak has subsided is about -134dBm so the low pass filter is providing about 39dB of attenuation outside the HF band. Before the peak at 31.5MHz and after 38.5MHz the shape of the curve is primarily dictated, not by the filter but by the SDR response outside the bandwidth setting. The amount of noise from the noise source is about 94-150= -56dBm and the filter is attenuating this by 94-133= 39dB. If the gain settings were higher then there's a chance that the SDR output will be non-linear, saturate at the maximum noise output and therefore the out of band response would be higher. Increasing gain to maximum would completely negate the effect if the filter. The second curve shows exactly the same set of conditions but with a 20dB attenuator between the noise source and the filter plus the Y-axis expanded (giving 112-148=-36dBm of noise and 147-112=35dB attenuation from the filter).

A curve (over 0-60MHz) of the same filter tested with the Rigol spectrum analyser .. click the picture to see more


Clearly the use of a noise source coupled with an SDR does produce useful results in terms of filter shape and frequencies but is not ideal when it comes to accurate measurements.. For those a tracking generator is far better than a noise source but of course a tracking generator plus a spectrum analyser is at least around £1500 compared with a half decent SDR and simple noise source which would set you back between £50 and £120. One of the reasons, in fact the true reason, for the poor performance of the noise source in determining filter characterisitics is its output across any given range of frequencies. My example manages about -56dBm of noise at 5MHz and the Nooelec maybe -67dBm. Compare this with a tracking generator which can run at 0dBm (see the picture immediately above). That is between 100 and 150dB above the natural baseline of the spectrum analyser so -60dB attenuation of filter is easily determined. Looking at the performance of an SDR it's baseline would need to be completely unaffected by the presence of a signal for 60dB attenuation to be measured. In fact if the noise source produced only around -60dBm at best at any given frequency the measurement of a filter would be tricky. As you can see above the 60dB attenuation of the low pass filter is only shown a little less than 40dB. In fact the total power from the noise source was measured at 1.4dBm spread over 10GHz whilst the tracking generator manages 0dBm at a single precise frequency. If the noise source was to compete with the latter its total output would need to be a kilowatt. Not a practical propostion for several reasons!

Before I wrap this part of the experiment up here are two more pictures. These are with the SDRPlay driven from the Nooelec noise source and then, without changing settings, from my noise source. Can you explain the differences? Click on these to see details.


 The next step in the development of the noise source was to see if a zener diode with a higher voltage rating would perform any better than that for 5.1 volts and, as the commercial Nooelec example uses a 7.5 volt zener, and as I have a large quantity of these left over from drive unit repairs, I tried an MMSZ7V5T1G. This change meant some modifications were required. Two additional AA cells were added and to prevent excessive voltage appearing at the two amplifier chips and I disconnected the 7.2 volt wire from the original zener diode and added a wire to the new 9.6 volt supply. To eliminate current drain in the off setting I changed the on/off switch to a double pole type. One pole switches the 9.6 volt supply from the zener diode to the charging circuit and the other switches the 7.2 volt supply to the amplifier circuit. I could have just increased the 180 ohm dropper to accommodate the extra 2.4 volts but that might add risk by a 9.6 volt pulse damaging the amplifiers whose maximum rating is 6 volts. I could add a 5.1 volt zener diode to the earthy side of the 180 ohm resistor and feed the whole circuit from 9.6 volts and keep the single pole on/off switch and that might be a further stage in development. Initially I need to see the effect of using the new zener diode.

I carried out the modification and a quick check with an SDR proved that the noise level did seem to be higher, however a test using the Rigol showed a clean 1.2GHz signal present at the output with noise above around 800MHz much reduced. This problem had been met previously although the oscillation had been lower, around 700MHz. That problem had been cleared up by adding extra decoupling capacitors so I tried that approach again. I found I could minimise feedback by adding a couple of capacitors but the level of the unwanted signal only dropped into the noise if I put my finger close the the first amplifier. Eventually, by trial and error the thing was almost tamed and the modified noise source was stable enough to be compared with the one using the 5.1 volt zener. Before modifying the circuit I'd taken some measurements at various frequencies and also duplicated these tests with the Nooelect noise source as well. The three sets of readings are now presented.

Read on to see the comparisons....


 Return to Reception