开云体育

ctrl + shift + ? for shortcuts
© 2025 开云体育
  1. HP-Agilent-Keysight-Equipment
  2. Topics

HP 346A Noise Source Below 10 MHz #file-notice

 

Sorry, I meant to also say that you can see that the trace noise is a bit more jaggy below about 3MHz and this is probably to be expected as the noise source ENR is starting to fall and the noise floor of the analyser is degrading by at least 1dB at the lower end of the frequency range. This is causing the gain and noise figure traces to look more jaggy. Averaging would hide this to some degree but I left it running without averaging to demonstrate this. So the system is struggling a bit more here.
 

I dug out my precision analyser preamp (homebrew) and used this instead of using the internal preamp in the PSA analyser. My preamp has ultra low VSWR and a 2.8dB noise figure across a few MHz up to about 1GHz. The combination of the low VSWR and lower noise figure should minimise the measurement uncertainty when using the 346A noise source. See below for a fresh measurement of the BFR91 amplifier using my preamp inline with the PSA analyser.
?
?
?
This has no averaging but you can see how much cleaner the traces are when using my external preamp. I also measured the amplifier on a 2 port VNA after an Ecal calibration at -25dBm source power.
?
?
You can see that the gain is very flat at 12.5dB across 300kHz to over 50MHz. This agrees with the gain measurement using the 346A noise source and the PSA analyser.
?
My external preamp was designed to have very low VSWR from 1MHz to 500MHz and a sub 3dB noise figure. It also has very flat gain across this range although this is less important.
?
?
You can see the input VSWR of my precision preamp is ultra low across 1MHz to 500MHz.??
?
?
?
This plot shows how flat the gain of the external precision preamp is. It was designed to act as a precision wideband amplifier to use for instrumentation and you can see there is no passband ripple and the VSWR is impressively low. The noise figure is a flat 2.8dB from a few MHz to well over 500MHz. The noise figure creeps up slightly down at 1MHz, but only by a small amount.?
?
It really helps to minimise measurement uncertainty when doing noise figure measurements with the 346A and PSA analyser.
?
Despite all this, I can't guarantee how accurate the subsequent noise figure results are below 10MHz for the BFR93 test amplifier, but if I assume the gain and noise figure is flat with frequency down to 2MHz, then it looks like I can assume high confidence in the results I've achieved. I've taken reasonable steps to minimise measurement uncertainty.
?
It would be interesting to see how others would approach this task and what results they achieved. Because the noise performance of the 346A isn't specified below 10MHz then it may be the case that there could be differences (between various 346A devices) in how flat the noise output is below 10MHz. I've got access to about eight 346A noise sources at work (but sadly no 346B) so I could maybe ask to borrow them and test them one by one? Some already have a fairly low ENR at 10MHz of only 5.0dB. I've been using them at work for about 20 years and I think they are are all 2004 (44) date code devices.
?
My 346A here at home has a 1993 date code.
?
?
?
?
?
?
?
?
?
 

It looks like my 346A can be used down to about 2MHz with good results. This may not be the case with all 346A noise sources though, especially if they are using different diodes for the newer noise sources.
?
Over the weekend, I could try and explore below 2MHz but I feel this is fairly pointless with my 346A as the ENR drops rapidly below 2MHz. However, it may be interesting to see how far down the PSA analyser can operate when I use the Noisecom noise source.
?
Both are meant to work down to 200kHz. Regardless of the result, I don't see this as a practical or reliable means to measure noise figure at such low frequencies.? I don't know if the BFR91 test amplifier can maintain a flat noise figure down below 2MHz. At some point it will start to show an increase in noise figure. I could also test a common gate JFET amplifier (with a BJT second stage) as this should hopefully show low 1/f noise and flat gain from maybe 30MHz down to very low frequencies. I just need to find it...
 

Now you have me curious; I just ran across a 346B in the lab at work this morning.? When I get back from lunch, I'll go check the date code.? ? Jim


On Fri, Oct 18, 2024 at 11:36 AM, jmr via groups.io
<jmrhzu@...> wrote:
It looks like my 346A can be used down to about 2MHz with good results. This may not be the case with all 346A noise sources though, especially if they are using different diodes for the newer noise sources.
?
Over the weekend, I could try and explore below 2MHz but I feel this is fairly pointless with my 346A as the ENR drops rapidly below 2MHz. However, it may be interesting to see how far down the PSA analyser can operate when I use the Noisecom noise source.
?
Both are meant to work down to 200kHz. Regardless of the result, I don't see this as a practical or reliable means to measure noise figure at such low frequencies.? I don't know if the BFR91 test amplifier can maintain a flat noise figure down below 2MHz. At some point it will start to show an increase in noise figure. I could also test a common gate JFET amplifier (with a BJT second stage) as this should hopefully show low 1/f noise and flat gain from maybe 30MHz down to very low frequencies. I just need to find it...
 

In case there are still any doubts about the low frequency VSWR of the 346A see below for a measurement of mine from 300kHz to 1GHz using an E5071B VNA and a 4431B-600006 Ecal module.
?
I know from experience that all of the 346A noise sources at work are very similar to this. I recall that the worst one had a VSWR of about 1.02:1 across 10MHz to about 500MHz. I don't think the VSWR was any worse than about 1.06:1 on any of them below 1MHz.
?
Below 1MHz, I think the VSWR will be flat at about 1.05:1 right down to VLF because the loss of the output attenuator will define this worst case VSWR at LF.
?
?
 

I’m probably restating the obvious, but more is more happening than just the VSWR. You have the DC block characteristics to consider, the noise diode mismatch uncertainty, and any effects related to the diode bias supply. A conical inductor might be good to 50GHz, but the might not work well at 1MHz, shorting some of the noise output to the bias supply (or allowing power supply, stray RF or other noise into the diode).
?
And all of that is on top of the attenuator not having constant attenuation vs frequency.
 

Yes, I had a digital noise generator in mind. It is a rather simple circuit.
The HP 3561A (and later models) dynamic signal analyzer has such a source, and uses it for various measurements. HP3561 only goes up to 100kHz, but with modern chips, 10MHz would be easy.
10MHz is in the "oscilloscope range", so you are not limited to GHz techniques. Of course, you could also make a (zener) diode noise source and calibrate it (after some amplification) with a HP3400, or a modern scope, with built in RMS calculation. Exact amounts of amplification up to 10MHz are also no problem with modern chips.
Most modern AWGs can also produce noise at least up to 10MHz. You would need some attenuation to get down to HP364 levels, but attenuation is also easy at 10MHz.
 

I agree that there are lots of contributors to overall uncertainty. However, I don't think a bias inductor is used to feed the diode with the correct current.?
?
The HPJournal from April 1983 gives the internal circuit for the 346B and this shows how the diode is matched using some resistors and a stub. I assume the 346A is the same but with a larger value attenuator at the output.
?
Looking at my old notes, I worked out that there is something like a 16dB attenuator at the tip of the 346A noise source and the blocking capacitor must be something like 1200pF.? I worked this out from a 1 port measurement of the 346A and I compared this to a model based on what is inside the 346 noise sources. Based on my findings, I think the coupling cap is something like 1200pF in my 346A but this may vary across device to device. I also worked out that the equivalent resistance of the diode and the matching resistors was about 44 ohms for my 346A.
 

The VSWR of a 346A below 10 MHz will necessarily be less than 1.052:1 since this corresponds exactly to the value of a 16 dB attenuator that equips all 346A. Contrary to what has been affirmed above, it is totally certain that the attenuator used is of the class of 8493C which guarantees it as totally flat over a few GHz and this from DC. Whatever will be the diode impedance, the difference in the measurement of the VSWR will hardly be seen. It would have to be exactly 50 ohms in order the VSWR will be excellent and of course it could not be possible due to the DC block.
?
?
 

Thanks. I have made several RF noise sources over the years and probably the best (and oldest) one is based on a Noisecom diode. This is a high output diode and so it is easy to achieve low VSWR with it because it requires lots of attenuation to get a reasonably low ENR. This one has removable SMA attenuators on it and if configured for ~13.6dB ENR it has an output VSWR better than 1.015:1 across LF through to about 1GHz.
?
I tried measuring the BFR91 test amplifier with it and got the result below. This is quite good although I wouldn't normally use this 13.6dB ENR noise source to measure a noise figure as low as 4dB. I would normally change the output attenuator to lower the ENR.
?
The Noisecom diode in this homebrew noise source is quite expensive but it usually works quite well. It isn't quite as stable as the 346A over time and temperature though. I think the spec for this Noisecom diode is a 0.01dB typical change in ENR per degC.
?
I have other noise sources that use really cheap diodes (less than $1) but these tend to drift at up to 0.1dB per degC. I'm not sure how stable they are over time and they are quite sensitive to changes in the bias current level as well. So the cheapo noise sources have to be used with some caution. I usually check them against a MMIC eval board I have here where I know the noise figure and gain.?
?
The plot below is for the Noisecom diode source measuring the BFR91 test amplifier and it goes down to 2MHz. I can explore below 2MHz but I'm not sure it will help the analysis of the 346A noise source.
?
?
?
?
 

I had some free time today to play a bit more with the 346A and the PSA analyser. This time I tested a common emitter amplifier using the classic 2SC3355 BJT from NEC. This was a very special BJT in its day that could deliver a sub 1dB noise figure when driven from a 50R source. I've designed quite a few amplifiers with this BJT and I'm lucky to have the manufacturer's non-linear model for it and this model has been proven many times to give realistic results for noise figure.
?
See the simulation below. This predicts a noise figure of about 0.8dB and this is about right. Because the input is unmatched, the input VSWR is about 8:1 and so this represents a significant mismatch at the amplifier input. This should be a good test for the 346A because of the mismatched input of this amplifier.
?
?
It's hard to read the text in the image above but the simulation shows a flat noise figure of 0.86dB across 2-20MHz. The gain is just over 21dB.
?
I built the amplifier and tested it for gain and input s11 on a VNA. The VNA source power had to be turned down really low to -40dBm to prevent any compression. This is a full two port test using an Ecal module to correct the ports of the VNA to be exactly 50 ohms. To compare against the PSA analyser the test must be done with a very small signal or subtle amounts of compression in the 2SC3355 could spoil the comparison.
?
?
The VNA shows a significant mismatch at the input (as expected) and the gain varies from 21.63dB at 2MHz to 21.46dB at 18MHz. The gain is very flat in other words and so is the noise figure.
?
See below for the result I got for noise figure and gain using the 346A noise source and the PSA analyser with my external preamp at the PSA input.
?
?
You can see it did a good job here. However, I think the ENR table isn't quite right yet. There does seem to be a bit of a stubborn bump at 5MHz so I think the ENR value at 5MHz needs to be turned down maybe 0.03dB or so. This plot is taken without averaging to show how smooth the results are even without averaging. It may be that the ENR value at 2MHz and 3MHz may need a tiny tweak as well, but I'm not too fussed about this.
?
This is the same ENR table as I used for the last tests. So it looks like the ENR table (across 2MHz to 10MHz) for this 346A noise source is close to being complete :)
?
?
?
?
?
?
?
?
?
?
 

FWIW, I couldn't find the date codes on the 346A and 346B at work yesterday, but the HP logo and Hughes Aircraft property tag date the 346A to sometime before the mid 1990s.? ?Likewise, the Agilent logo and Raytheon property tag put the 346B at probably early 2000s.? I assume one would have to disassemble either to find the date code.? That would be a CLM (Career-Limiting Move) where I work....? ? ? Jim?


On Sat, Oct 19, 2024 at 12:54 PM, jmr via groups.io
<jmrhzu@...> wrote:
I had some free time today to play a bit more with the 346A and the PSA analyser. This time I tested a common emitter amplifier using the classic 2SC3355 BJT from NEC. This was a very special BJT in its day that could deliver a sub 1dB noise figure when driven from a 50R source. I've designed quite a few amplifiers with this BJT and I'm lucky to have the manufacturer's non-linear model for it and this model has been proven many times to give realistic results for noise figure.
?
See the simulation below. This predicts a noise figure of about 0.8dB and this is about right. Because the input is unmatched, the input VSWR is about 8:1 and so this represents a significant mismatch at the amplifier input. This should be a good test for the 346A because of the mismatched input of this amplifier.
?
?
It's hard to read the text in the image above but the simulation shows a flat noise figure of 0.86dB across 2-20MHz. The gain is just over 21dB.
?
I built the amplifier and tested it for gain and input s11 on a VNA. The VNA source power had to be turned down really low to -40dBm to prevent any compression. This is a full two port test using an Ecal module to correct the ports of the VNA to be exactly 50 ohms. To compare against the PSA analyser the test must be done with a very small signal or subtle amounts of compression in the 2SC3355 could spoil the comparison.
?
?
The VNA shows a significant mismatch at the input (as expected) and the gain varies from 21.63dB at 2MHz to 21.46dB at 18MHz. The gain is very flat in other words and so is the noise figure.
?
See below for the result I got for noise figure and gain using the 346A noise source and the PSA analyser with my external preamp at the PSA input.
?
?
You can see it did a good job here. However, I think the ENR table isn't quite right yet. There does seem to be a bit of a stubborn bump at 5MHz so I think the ENR value at 5MHz needs to be turned down maybe 0.03dB or so. This plot is taken without averaging to show how smooth the results are even without averaging. It may be that the ENR value at 2MHz and 3MHz may need a tiny tweak as well, but I'm not too fussed about this.
?
This is the same ENR table as I used for the last tests. So it looks like the ENR table (across 2MHz to 10MHz) for this 346A noise source is close to being complete :)
?
?
?
?
?
?
?
?
?
?
 

Thanks. I don't know if this is of any interest, but see below for my original attempt at reverse engineering what attenuator and blocking cap was used in my 346A noise source. The circuit was based on the info in the April 1983 HP Journal. Sadly the HP Journal didn't provide the value of the attenuator or the coupling cap for the 346A so that was why I had to try and work it out using the simulation below. I compared against a VNA measurement of my own HP 346A noise source.
?
I ended up with about 1200pF for the coupling cap and about 16.2dB for the attenuator. This gave the best curve fit for the model.
?
?
?
?
?
?
If you look closely, there are two VSWR plots and two Z complex plots and I adjusted the blocking cap and the attenuation value to try and get the real and imaginary parts of the impedance to overlay as accurately as possible when compared to a VNA measurement of a real 346A noise source. I ended up with a diode resistance of just 15 ohms in the hot state. This seems really low but it is the only way to get a match for the real and imaginary parts of the impedance across LF through to about 20MHz.
The traces almost overlay perfectly below 10MHz.
?
You can see that the power transfer suggests a relative loss of only about 0.25dB between 10MHz and 2MHz although this assumes the noise output is flat with frequency. This is partly why my first attempt at an ENR table wasn't quite right at the 2MHz and 3MHz cal points. This agreed with my first attempt at measuring the output noise. The value of the coupling cap is quite critical to the level response below 3MHz so hopefully this cap is stable over time and temperature. I hope this simulation is interesting. I'm not sure how much value it adds to the discussion though.
 

As I mentioned earlier, you can work out the resistive equivalent of the resistive matching network and the diode when it is in the hot state. Just put the 15R (hot diode resistance) in parallel with the 22R? and then put this in series with 34R. You end up with about 43 ohms resistance.
?
I suppose this ought to be 50 ohms but maybe it is a compromise because the ESR of the blocking cap and the connection path will rise with frequency (could be several ohms by 18GHz) and also you have to factor in the impact of the open stub matching section (not shown) up at higher frequencies. The open stub shouldn't affect the simulation below 20MHz so I didn't include the 30 ohm resistor or the stub. Only the original designers at HP will know for certain why these values for the matching resistors were chosen.?
 

Another way to expand on the 2SC3355 test is to add a 10dB attenuator at the input. I used a decent Suhner attenuator with ultra low VSWR and it has 10.10dB attenuation at 10MHz when measured on a VNA.
?
When I add this at the input of the 2SC3355 amplifier the gain drops by 10.1dB and the noise figure goes up by about 10.1dB across 2-20MHz. The Y factor becomes really low when using a 5.5dB ENR source with a DUT noise figure of about 11dB but the result was still quite good. I think the bump at 5MHz is still there but it's lost in the trace noise on this particular sweep below. The system is struggling a bit here because the Y factor is quite low with the 10dB attenuator inline.
?
?
Replacing the 10dB attenuator with a precision 3.00dB attenuator at the input gave the response below:
?
?
I do think there is a very slight issue with the ENR table at 5MHz and I could probably improve things if I used averaging to find out how much to tweak it by. However, I'm not sure it's worth the effort because I'm unlikely to ever do any critical noise figure measurements down at 5MHz :)
?
These test results look good to me and it may be the case that the ENR down below 10MHz will be a bit more drifty over time and temperature anyway. So it might be a wasted effort to try and improve it any further.
?
?
?
 

Here's another amplifier that should have flat noise figure down to below 2MHz. This is a high reverse isolation amplifier using two dc coupled BFR91 BJTs. It has ultra low port VSWR and the noise figure is about 8.3dB up at RF.
It works up to several hundred MHz and has about 13dB gain. I've used this amplifier many times in various test systems here so I know it well.
?
It is possible to reduce the noise figure by several dB by adding a choke to one of the bias resistors but I've removed the choke to make sure the noise figure and gain is very flat to low frequencies.
?
In the plot below I used four averages and the subtle bump at 5MHz is still there. It's much easier for me to see it on a sweep by sweep basis and I can only post up single sweep responses here.?
?
The main thing to note is the noise figure is once again flat vs frequency and if the 346A is accurate at 10MHz where it shows a noise figure of 8.3dB and a gain of 12.99dB I think it's reasonable to suggest I've also made a good noise measurement at 2MHz as this BJT amplifier should have flat noise right down to low frequencies.?
?
I have also tested a few InGaP MMICs and these don't show a flat noise response at low frequencies. They have some 1/f noise and this can slightly affect the noise figure at 2MHz.
?
 

To investigate this further, I measured four different 346A noise sources on an E5080A VNA. These are all made in 2004 and the serial numbers are quite close together. They are all within 150 of each other.
?
The reactive part of the impedance looks quite similar but the resistive part is different and you can see some spread in VSWR. I suppose it's possible they may have been stressed by reverse power at some point but this is unlikely.
?
This might represent the typical spread in VSWR seen with these noise sources. It may be that one or more of the resistors in the matching section is a select on test device with the aim of achieving low VSWR and best noise response out to 18GHz.
I'm not sure how much spread to expect in the attenuation value of the 16dB attenuator at the output port of the noise source.
?
?
?
 

I'm not sure if it's wise to put up my final ENR table for my 346A (HP 346A made in 1993) as there will be spreads between these devices. So it is going to be of limited use to anyone else. Also, I don't know how stable the noise output will be below 10MHz over time and temperature. So it might be quite different in a few months or years.
?
But here it is anyway...
?
2MHz? 4.65dB
3MHz 5.05dB
5MHz? 5.47dB
10MHz? 5.59dB This is the factory calibrated value stamped on it at 10MHz.
?
So it looks like the ENR is just under 1dB lower at 2MHz compared to the factory value stamped on it at 10MHz. This is about 0.65dB lower than my initial estimate at 2MHz but I didn't spend much time on it back in 2019.?
?
A lot depends on the value of the blocking cap after the attenuator and also the noise output of the diode at low frequencies. I really don't know how stable it will be over time. Above 10MHz it is remarkably stable because I've had consistent results from it for many years and I can leave it for several hours on the analyser without seeing any drift.
?
I also checked out the Noisecom diode on the RSA3408A and it does appear to be flat to almost 200kHz. There is some rolloff below about 500kHz but not much. My other homebrew noise source is the one with a Schottky diode. This is a low barrier device with a guard ring and it gives remarkably flat noise out to about 5GHz. It is also essentially flat down to about 500kHz. It is a bit drifty with temperature changes though...?
?
?
 

Arbitrary generator! You could generate any noise distribution you need in python and load it. Very long sequences are possible. Ref. Mersenne Twister and the like.?
 

I guess the challenge would then be to be able to accurately attenuate the arb output down to a level similar to an HP346A or maybe just the HP346B. I'm sure it can be done but it will need some care.
?
I've now got two more 346A noise sources here. Both were made in 2004 and both have current cal certificates from Keysight. I did a quick look on the Tek RSA3408A and they both look to be very similar to my 346A in terms of how much the noise rolls off below about 3MHz down to 1MHz. I'll do some more tests over the weekend.

21 - 40 of 42