Is this a gate/drain capacitance problem?

I have a noise generator consisting of a reversed bias "zener" diode (8.2V), a 28V supply and a resistor of about 2 kOhm. There is a commerical bias T that has and L and C in it.

As the 28V is witched on/off, so the diode avalances or not, and so noise is produced. So the circuit is something like this, although the L & C are are bit more than just simple compoenents, as this bias T worked from 20 kHz to 8GHz.

+28V | | R | | L | | Zener ---||-------noise out. | | ground

I wanted to add a method of TTL control of the noise. So stuck the drain of an n-channel fet between the R and L, and grounded the source. As the gate is driven, it basically crowbars the supply to the zener. That seems to work OK.

What I find odd is that if there is no 28V supply at all, but the gate of the FET is drive, so "spikes" of noise appear on the output as the TTL signal changes state.

Do you think I am just injecting charge ito the gate, some of that gets to the drain, and so the drain develops a voltage on it, that is sufficient to cause the diode to generate noise? I can't see how, as the diode will not generate much noise unless it avalacnhes, and the TTL signal is less than the breakdown voltage of the diode. Perhaps the L's that are around are messing things up.

I'm using a power FET in a TO-220 case, simply because that was all I had, but I'm wondering if swapping to a low power FET will cure this?

In normal operation, there is no reason to use the TTL drive unless there is the 28V supply present. But I'm concerned that since I can see this effect with no 28V supply, it is probably there (but not so visable) when the supply is there.

It'll have a lower drain-gate capacitance. Do you get the same thing if you replace the FET with a small capacitor?

Two alternatives that spring to mind:

1. PNP BJT switching +28V on/off
2. NPN BJT with base to +5V; emitter (via series resistor) is TTL input; collector to R / L junction

Hi, Dave,

Not terribly elegant. Apart from your problem of spikes, both with and without 28V, there are issues of quiescent power dissipation.

With 28V present, the spike will be much bigger because you're essentially changing the current through the Zener.

Does your noise generator have a specification? You have done it the cheap and cheerful way. Does it need to be any better?

Yes, a small-signal FET will clearly have lower gate-drain capacitance, but not zero. If the spike amplitude is still too high, then you might have to go to a reed relay.

Cheers,

Zigoteau.

Which is next to zero. I'm dropping about 20 V across a 2.1 k resistor, which is < 200mW.

Spike I would like to reduce.

How would you do it in a more elegant way?

But that will happen anyway. Normally the 28V line is switched. In that case I am always going to have to fight the V = L dI/dt.

I have a commercial (10 MHz to 18 GHz) noise source that works this way, but the noise is produces is too low (its about 15 dB above thermal noise). This is not enough for what I need.

Not a formal one. I am doing this for a lab experiment, and to have a source around of high noise source I need one again.

Since I intend keeping it for a while, and perhaps use with a lock-in amplifier, which will be messed up to a small extent by the spikes, I would rather get rid of them if I can.

Applying the TTL without the 28V is not a mode it will be used in. That is for sure.

But it will be used in

a) 28 V switched on/off at about 10 Hz by a commercial noise figure meter, with the TTL open (i.e. the 28V works to switch the noise on/off).

I assume the noise figure meter gives the transients before it makes measurements.

b) Me applying 28V constantly and using the TTL drive from perhaps a lock-in amplifier for some tests. The TTL was very much an afterthought I added "just in case I have a need for it".

But I am concerned the noise the TTL it is causing. This might be an issue if it caused a massive spike when it switched the 28V supply. With zero current in the diode, the diode will generate the same noise power as that of a resistor at room temperature (about 295K). With the 28V on, it will generate the same noise power as a resistor about about

100,000K. I hope the transients are not causing it to generate the same noise power as a resistor at 1,000,000 K at the time of the transient.

I realise that. It might be sensible to go to a smaller device. I have some microwave devices around. I might look at using one of those.

I want to avoid a mechanical unit. The delays induced by them will probably mess things up a lot more than the transients.

Is that any better than just switching the 28V on/off? This wsll under some circumsttances be used with a HP 8920A noise figure meter

which I assume will switch the 28V on, wait for any transients to settle, make a measurement, switch the 28V off, waith for transients to settle etc.

The noise figure meter will supply its own switched 28V supply.

However, for some other uses, a fixed 28V supply and a TTL annd/or CMOS compatible input would be nice.

Is that likely to be better? There is no 5V inside, but I could derrive one. from 28V.

Hi Dave,

I am not familiar with handling 18GHz. Do you need special transmission line techniques? I can see that components might not work the same way at 18 GHz as they do at 1 kHz, and this explains why you must switch the zener diode on and off, rather than switching after removal of the DC component.

It sounds as if you do not really need to have a transition between noise-on and noise-off faster than a millisecond, and it also sounds as if the source does not have to produce noise all the way down to baseband. The easiest way to get rid of the switching transient is to filter your TTL output to the FET gate so that it is no faster than you need, perhaps with an RC filter.

Cheers,

Zigoteau.

Consider leaving the noise generator on all of the time. Then add a PIN RF switch in series with the output of the generator.

Hi, Dave,

I don't want to be rude, but I think you are being rather picky, and that it would help if you first took the time to understand the subject.

If you remember, you sent a related post to sci.physics, in which you also said P=kTB. I responded that, while it contained the ghost of a true formula, it was not completely correct. Your formula assumes no reflection, i.e. that your source and load impedances are perfectly matched to your transmission line, and are the circuit equivalent of black bodies. I suspect that the calibrated noise source bandwidth of

30 kHz-8 GHz you are asking for may be technically impossible, and in any case requires much better facilities than you are likely to have in your garage.

I don't know why you are interested in noise intensities much higher than thermal. Surely you will only want to make measurements on low-noise amplifiers? Since the best amplifiers approach the thermally-imposed limits, a resistor or diode should do what you want. AFAIK the spectrum of the noise from a zener diode depends on the parame of the particular diode, so is not as reliable as a noise reference as the other two.

Since your HP unit has just arrived, why not try it with your current circuit. As you say, it probably waits for a while until switching transients have died down before starting serious measurements.

Good luck with your project.

Cheers,

Zigoteau.

I bascially want this to work wide-band - 30 kHz at and 8 GHz might be the 3dB points when I receplace the standard zener 8.2V zener diode for a diode designed for generating noise, it should go to around 8GHz.

Whilst PIN diodes can switch at 8 GHz, they are limited at the lower end by the carrier lifetime. I don't think you can switch below 10MHz or so.

GaAS switches might work, but will introduce a lot of extra complexity and cost.

However, the biggest problem with these is going to be isolation. When the noise is "off" it must really be off. I am relyking on knowing the noise power in the off position to be

P = k T B (k=Boltamak, T in temperature in Kelvin and B Bandwidth in Hz).

For this reason I have a temperature sensor too, so I can measure compute the off power. The leakage trhough the switch will mean the off power is > k T B.

Yes, at 18GHz, you would use transmission line techniques. The commercial source will definitely use it.

My homemade source is part homemade, and part commercial, in that the biasing arrangement (the bias T, consisting conceptually of just and L and C, but in practice a lot more), is a commercial bias T, that was lying around. That is speced from 30 kHz to 8 GHz. At the time purchased, it was state of the art, but no longer is.

The commercial noise meters (such as the HP 8970A I am expecting will arrive today) will I am sure give any transients time to settle before making a measurement.

I want it as low as possible. The biasing arrangement for the diode will limit the -3 dB point to 30 kHz at the low end.

Actually, thinking about it, I think that will make matters a lot worst! The reason being, that the amount of noise produced by the diode at low frequencies just at the point of avalanching is very large indeed. The diode is not operated at that voltage, as it is too unstable, but keeping the diode voltage in an intermediate state for any longer than necessary will make matters very poor.

If you look at the noise output on a spectrum analyszer, whilst adjusting the diode current with a DC power supply, you will see what I mean.

I have looked it this in some detail, so think I have quite a good understanding. I've dug out the original paper on this by Nyquist, where I see kTB is just an approximation.

I've also found later work that shows the noise is not 0 at 0 kelvin, but equal to h f / 2 - not that it makes any difference to me.

I can't recall sending one to sci.physics, but I have sent some around on this issue to several places, so I could well believe I did.

I'm not doing this in a garage, but a university research lab. That said, some of the equipment (HP niose figure meter and commerical noise source) have been purchased by me personally.

I don't need it calibrated from 30kHz to 8GHz. But they are the expected

3dB points. Assuming the noise diode is flatish from 10 Hz to 8GHz (it is specified for use over that range), the bias T has 3dB points of 30kHz and 8GHz, I should get something approaching a reasonably flat response over that range once I get the NoiseCom noise diode, which is on order.

No,

I have a particular interest in making comparative noise measurements on a system that generates considerable noise and has very considerable loss. I might well need to amplify the noise and turn the amplifier on/off with the noise diode.

I have on order a diode characterised for noise. It should have an ENR between 30 and 35dB over the range 10Hz to 8GHz.

However, I am also intersted in producing a calibrated low-noise device, so are in fact building the one device, but will change the internal attenuator from 3 dB to 30 dB once I have finished with the high-noise measurements.

I don't believe the poor mismatch with a 3dB attenuator will present me a problem, since at the high noise levels, I am not going to be able to make calibrated measurements. This is not a simple RF amplifier - I'd rather not go into much detail on this, as it will be the subject of a scientific paper I hope to submit for publication soon.

The HP unit has not arrived - something I am getting a bit concerned about now. USPS tracking shows it has been delivered, but I believe that only works in the USA, and does not follow progress once in the UK.

I am trying to build the noise source to satisfy 4 jobs, with the only change being in the value of the attenuator on the output of the noise source.

1) Lots of noise - even 33 dB above thermal may not be sufficient. Poor match will be irrelevant. 2) Little noise for measurement of low noise devices. For this, I will add a 30 dB attenuator internal to the source. This is the reason I have the temperature sensor too. 3) Use with the noise figure meter, where I suspect the meter will wait for transients to die down. This will be both with the high and low noise measurements. 4) Possible use with a lock-in amplifier, where the transients will screw the results to a certain extent. Probably not a lot, but it will mean using longer time constants on the filter to ensure the transient has little effect.

I hope you can see why some of my design ideas may seem excessive. I'm trying to build one source that should only need a change of output attenuator to do some different tasks.

The fact that that temperature sensor will not be used at high noise measurements, and that the noise meter probably waits for transients to settle, does not mean I can miss out the temperature sensor or ignore the transients.

Thank you.

I think part (perhaps all) of the problem is that as the diode is switched on or off, there is a point at which the voltage is such that it jut avalanches. I know that produces noise some 10-15 dB higher than when the voltage is increased a few hundred mV. When the diode is just avalanching, the noise at low frequencies is probably 40~45 dB more than thermal. Unfortunately, it is not stable at that point.

Hi Dave,

Thanks for the additional details of what you are trying to do

With all due respect, you're not there yet.

Absolute zero temperature does not exist, but 4.2K does, and the simple noise formula needs amending in that case. I'm afraid the correct version is not " hf/2". Have you downloaded the paper I found for you?

publication/julien.pdf

Actually it was sci.physics.electromag. You talked about "the noise power from a conductor", which sounded more like garage-level than university research. I cannot believe that mismatch does not matter.

Cheers,

Zigoteau.