|
Has anyone investigated the ENR and ENR uncertainty of the HP 346A noise source at frequencies below 10 MHz? Same question for the HP 346B?
?
If so, what was found?
?
Thanks!
--
Whitham D. Reeve
Anchorage, Alaska USA
---
|
346 noise sources are not really usable below 10 MHz. The match becomes very bad due to a d.c. block between the noise diode and output attenuator. So, even if you characterized the noise that does make it through the capacitor, you would be dealing with a bad return loss.
?
Vladan
|
If you need a very low band source below 10 MHz, you could conceivably use one of these heads with a high impedance buffer amplifier. That would take care of the LF roll-off from the coupling cap. You'd need to account for the gain scaling and additional noise of the amplifier of course, but it should be straightforward. Down at the low frequencies you won't be worried about impedance matching at the input, just straight voltage gain (like unity), and let the output take care of matching to 50 ohms or whatever.
?
A simple emitter follower or JFET follower or totem-pole (or even an opamp) should easily perform the Z transformation over reasonable bandwidth. Depending on circuit design, you'll still probably have an output coupling cap, but you can make it as big as you want. With opamp circuits, you can go with DC output coupling to get way down, leaving only the input side to limit the LF response.
?
With these 34X noise sources, I think it's only one stage, with a noise (avalanche) diode and coupling network, so probably only one dominant capacitor for the low end.
?
Ed
|
Over my career, I've used a lot of HP 346A noise sources (5.5dB ENR) but only a few examples of the 346B (15dB ENR).
?
In my experience, the input VSWR of the 346A (in both hot and cold states) is usually very low across 10MHz to a couple of GHz. A typical VSWR figure might be 1.02:1 across this frequency range. A good example might achieve a VSWR of < 1.01:1 at 10MHz. At frequencies below 10MHz, the VSWR will gradually degrade, but I don't think it will get much worse than about 1.06:1 even below 1MHz.?
?
Probably the biggest issue with the 346A will be finding an instrument that formally supports it below 10MHz. In terms of noise figure measuring performance, the performance of a typical analyser will degrade quite a bit below 10MHz in various ways. This makes it difficult to support a noise source with a low (5.5dB) ENR like the 346A. You may be better off using a 346B with (say) a 6dB attenuator fitted to it to lower the 15dB ENR a bit and also improve the source match. This might achieve a better compromise for use below 10MHz.
?
It's a while since I used a 346B so I've forgotten the typical input VSWR vs frequency. The VSWR will be higher for the 346B because it has less internal attenuation after the noise diode. One option is to fit a low Q matching network at the input of the 346B as well as an external attenuator. This would hopefully deliver a decent ENR of maybe 9dB as well as providing a good source match below 10MHz. This should improve the overall uncertainty when using the noise source although you would have to find a way to determine the ENR of such a setup when used below 10MHz.
|
I do not know the specific application, or I missed something, but below 10 MHz noise figures are generally spoken not so crytical an usually quite high. Couldn’t you just go for the 3dB method? As said above the HP346A/B/C needs dedicated equipment that normally does not operate below 10 MHz.? Best regards, Harke
(Send from my iPad4)
toggle quoted message
Show quoted text
Op 15 okt 2024 om 18:21 heeft jmr via groups.io <jmrhzu@...> het volgende geschreven:
? Over my career, I've used a lot of HP 346A noise sources (5.5dB ENR) but only a few examples of the 346B (15dB ENR).
?
In my experience, the input VSWR of the 346A (in both hot and cold states) is usually very low across 10MHz to a couple of GHz. A typical VSWR figure might be 1.02:1 across this frequency range. A good example might achieve a VSWR of < 1.01:1 at 10MHz. At frequencies below 10MHz, the VSWR will gradually degrade, but I don't think it will get much worse than about 1.06:1 even below 1MHz.?
?
Probably the biggest issue with the 346A will be finding an instrument that formally supports it below 10MHz. In terms of noise figure measuring performance, the performance of a typical analyser will degrade quite a bit below 10MHz in various ways. This makes it difficult to support a noise source with a low (5.5dB) ENR like the 346A. You may be better off using a 346B with (say) a 6dB attenuator fitted to it to lower the 15dB ENR a bit and also improve the source match. This might achieve a better compromise for use below 10MHz.
?
It's a while since I used a 346B so I've forgotten the typical input VSWR vs frequency. The VSWR will be higher for the 346B because it has less internal attenuation after the noise diode. One option is to fit a low Q matching network at the input of the 346B as well as an external attenuator. This would hopefully deliver a decent ENR of maybe 9dB as well as providing a good source match below 10MHz. This should improve the overall uncertainty when using the noise source although you would have to find a way to determine the ENR of such a setup when used below 10MHz.
|
I've never attempted to use one at the bottom end of the spec'd frequency range, but based on my (limited) experience with homebrew noise sources, I'd question the stability of the ENR down there. Avalanching gets you a high ENR without operating at incandescent temperatures, but it is flaky. The 1/f corners are high and not particularly stable. I assume that the 346x uses buried devices to moderate the instability, but there's only so much that one can do.
Whether any of that matters depends, of course, on what you need out of the NF measurement. How accurately, and over what frequency range, do you want to measure NF? For terrestrial wireless, for instance, ambient noise typically sdominate over intrinsic thermal noise below about 20-30 MHz. Making accurate NF measurements at 1 MHz, say, isn't particularly important if the idea is to infer the sensitivity of AM radios, for example.
And if you only want to measure NF below tens of MHz, there are several other options for doing so.
-- Cheers
Tom
--
Prof. Thomas H. Lee
Allen Ctr., Rm. 205
420 Via Palou Mall
Stanford University
Stanford, CA 94305-4070
toggle quoted message
Show quoted text
On 10/13/2024 4:46 PM, WReeve wrote: Has anyone investigated the ENR and ENR uncertainty of the HP 346A noise source at frequencies below 10 MHz? Same question for the HP 346B?
|
Looking back through my old notes, the ENR at 1MHz was lower than 5.20dB so this may be a typo on my ENR table stored in the analyser. However, I possibly boosted it to prevent 'pulling' of the NF results when operating just below 2MHz. My notes suggest that the ENR drops quite rapidly below about 1.5MHz. I should probably have added an extra ENR table entry at 1.5MHz because of this.
?
I wouldn't trust it below about 2MHz anyway. There are too many contributors to overall uncertainty. The analyser uncertainty will be quite significant here although the use of an external LNA helps in this respect. Not many modern analysers from HPAK are going to be usable below 10MHz unless an external LNA is used. The noise floor and VSWR is going to degrade below 10MHz with the internal factory preamp enabled. My analyser manages a noise floor of about -170dBm/Hz down to a few MHz with my external preamp fitted and the preamp input VSWR is about the same or better than the 346A noise source VSWR at these sub 10MHz frequencies.
?
?
|
JMR, your numbers didn't sound right to me, so I made the following measurements on a 346B. The OP is curiously absent, so we may never know what he wants to do.
?
Vladan
?
|
Hi Tom,
?
Correct, things become unpredictable at lower frequencies. The original 346 diode was a special design in that, among other things, it had a p-doped guard ring to control the avalanche area as tightly as possible. The diode is in a hermetically sealed assembly and it was eutectically attached to the package to stabilize the temperature. Later, when HP obsoleted the Si process, the diodes came from Metelics, which unfortunately went downhill after being purchased by a conglomerate. I am not sure about the differences between old HP and Metelics diodes.
?
Vladan
|
Just to clarify, the thread title does ask about the HP 346A and so I gave estimated VSWR data for the HP 346A. The 346A has much lower VSWR than the 346B. You can typically expect to see less than 1.02:1 VSWR at 10MHz with the 346A and this will degrade to about 1.05:1 by about 300kHz. It might degrade to 1.06:1 by 100kHz.
?
I did also say I couldn't remember how much worse the 346B VSWR was. I would guess that the 346B has maybe 9dB less attenuation after the diode because it has a higher ENR at its output. So the VSWR will obviously be worse than the 346A in this respect.
?
?
?
?
|
You are correct, JMR. The 346A is a lot better below 10 MHz than the 346B.
?
Vladan
|
Thanks. Looking at your other plot, I think your spectrum analyser plot is a plot of noise output vs frequency. It looks like your noise source has a steeper noise reduction than mine as the frequency is reduced. It may be that your noise source has a different capacitance value for the dc blocking cap inside it?
?
I can try and come up with a better way of measuring mine to try and characterise it better. I didn't spend much time on this last time. I don't think it will reduce by 2dB by 2MHz and by 4dB at 1MHz though.?
?
I've also got a couple of homebrew noise sources here. One contains a decent Noisecom noise diode and the other uses a high frequency diode. I think it's a Schottky diode but I can't remember now. It works very well up to several GHz. The Noisecom source is only really good to about 1300MHz although it is supposed to work down to 200kHz according to the datasheet.
|
I'm probably going to do it the same way as last time using an old Tek RTSA as the noise measuring receiver.? This analyser has a low noise floor and a very flat response and a very flat noise floor right down to audio frequencies. The technique last time was to use a BJT based wideband 50R amplifier to drive an InGaP MMIC to act as a preamp.?
The assumption is that the noise figure of the BJT stage (BFR91 based) should be consistent from about 40MHz down to maybe 2MHz or even lower. So hopefully, the preamp will not contribute any slope to the noise response seen on the analyser at lower frequencies.
?
The RTSA frequency response and noise floor is very flat when used below 40MHz because it acts as a baseband analyser below 40MHz. However, I am going to try and improve the BJT stage to improve the input and output return loss. I can also tweak the InGaP MMIC board to improve the frequency response below 10MHz to make it ultra flat. It's going to be quite difficulat to manage all the contributors to uncertainty, but this is realistically the only way I can do this here at home.
?
If the ENR of the noise source was totally flat then the noise level seen on the analyser should rise but still be flat once the noise source was set to the hot state. The last time I did this I saw a slight droop down in noise level between 5MHz and 2MHz. Hopefully, I can modify the BJT amp and the MMIC to minimise mismatch uncertainty and also improve the frequency response. This should provide a more reliable result.? I probably won't be able to do all this until the weekend though.
?
It's important to use a preamp with low 1/f noise and so I used the BJT stage for the first stage of the preamp.?
|
?
I dug out and tried the original external preamp on the RTSA (Tek RSA3408A) and then fitted the 346A at the input. I then flipped between hot and cold states. See the screenshots below. You can see there is a flat response from 10MHz down to maybe 3MHz and then there is some obvious loss in noise level below 2MHz. This is essentially the same test I did a few years ago.
?
This is about the same as I saw when I did this last time. I probably also (further) tweaked the ENR table in the noise figure analyser menu after I did all this. I can't remember now as I did it quite a few years ago when I first got the noise source and the associated analyser that can measure noise figure. I think this was back in late 2019.
?
The first plot shows the cold state and the noise floor is really flat. The noise floor is just as flat for just the raw analyser (without the preamp) but obviously at a much lower level. The typical noise floor is about 20dB lower without the preamp.
?
The second plot shows what happens when I turn on the HP 346A. It looks like the droop in noise level starts below about 3MHz and this gets worse by 1MHz.
?
I just spotted that I should have calibrated the analyser just before running these plots as it's possible to reduce the dc spike with a calibration. However, this doesn't really affect the results.
?
If anyone is interested, I can share the design of the BJT stage although it might be better to wait until I tweak it for (very) low port VSWR.
?
The MMIC stage is just a Minicircuits eval board but I need to change the biasing choke and optimise the biasing for best performance.
?
It may be the case that the result will change once I tweak the preamp for lower VSWR and a flatter response but I doubt it will change a lot.
?
The input VSWR of the analyser is impressively low even when set to its most sensitive range because it is set to act as a baseband analyser in the plots below. So it is a bit like a 40MHz soundcard with a 50R input.
?
?
?
?
|
Don't pay too much attention to the 0.43dB value seen on the delta marker in the plots as this jumps around quite a bit even with lots of averaging.? However, there's about a 3.5dB difference (Y factor) between hot and cold so the ENR is going to be in the ballpark of 0.7dB lower at 2MHz?
?
On my old notes I only saw about 0.2dB noise droop at 2MHz but I probably didn't take that much time or care over this. I may have used a different INGaP MMIC as well. I'll try and sort out the preamp over the weekend. The VSWR and frequency response is already quite good but I can make it even better.
|
I think at these low frequencies, a digital source would be much better? A linear feedback shift register with 100MHz clock with a lot of post attenuation would give a perfect spectrum at 10 MHz. A simple CPLD could be programmed, or maybe even 74AC logic would do. You can make the register arbitrarily long, to avoid periodicity.
|
Uh, Marko, you appear to be responding to the wrong topic.? No offense, just letting you know.? ? ? ?Jim Ford? ? Laguna Hills, California, USA?
On Wed, Oct 16, 2024 at 9:50 PM, Marko Cebokli <s57uuu@...> wrote: I think at these low frequencies, a digital source would be much better? A linear feedback shift register with 100MHz clock with a lot of post attenuation would give a perfect spectrum at 10 MHz. A simple CPLD could be programmed, or maybe even 74AC logic would do. You can make the register arbitrarily long, to avoid periodicity.
|
I think Marko is referring to a 'digital' noise source. I've got various digital noise sources here and some are built into vector signal generators.?
?
For example, an additive white gaussian noise generator (AWGN) can generate noise quite accurately. These can be homebrew or you can buy them. The AWGN feature in my sig gens can be set to have a random seed and then it produces a long random sequence of noise. It does repeat itself after a while but the repeat rate can be controlled in the menu. It's possible to define the bandwidth and the power level of the noise. A decent attenuator would be needed in order to get the level down to a very low level. A typical RF sig gen can do this with it's internal step attenuator.
?
I think the main niggle with using the 346A below 10MHz is the low ENR it offers and this gets lower at frequencies below about 2 or 3 MHz. The noise floor of a typical RF/Microwave spectrum analyser will gradually rise below 10MHz and this increases the overall measurement uncertainty when trying to measure noise figure with a low ENR noise source. The noise figure analysers are only spec'd down to 10MHz and I'm only aware of one HPAK RF/microwave spectrum analyser that can work with the 346A below 10MHz with reasonably low uncertainty. Even then, it isn't an ideal combination.
?
?
|
My mistake.? ?Thanks for the correction, jmr.? Hmmm...? digital AWGN generation, eh?? Sounds like a fun project to be filed away in the Someday File.? Jim
On Thu, Oct 17, 2024 at 9:44 AM, jmr via groups.io <jmrhzu@...> wrote: I think Marko is referring to a 'digital' noise source. I've got various digital noise sources here and some are built into vector signal generators.?
?
For example, an additive white gaussian noise generator (AWGN) can generate noise quite accurately. These can be homebrew or you can buy them. The AWGN feature in my sig gens can be set to have a random seed and then it produces a long random sequence of noise. It does repeat itself after a while but the repeat rate can be controlled in the menu. It's possible to define the bandwidth and the power level of the noise. A decent attenuator would be needed in order to get the level down to a very low level. A typical RF sig gen can do this with it's internal step attenuator.
?
I think the main niggle with using the 346A below 10MHz is the low ENR it offers and this gets lower at frequencies below about 2 or 3 MHz. The noise floor of a typical RF/Microwave spectrum analyser will gradually rise below 10MHz and this increases the overall measurement uncertainty when trying to measure noise figure with a low ENR noise source. The noise figure analysers are only spec'd down to 10MHz and I'm only aware of one HPAK RF/microwave spectrum analyser that can work with the 346A below 10MHz with reasonably low uncertainty. Even then, it isn't an ideal combination.
?
?
|
Yes, I think the AWGN technique gets used a lot, both with real hardware and with simulations (eg Matlab or Octave etc).
?
I had a go at measuring the BFR91 preamp section on an old 26.5GHz Agilent PSA analyser today (with my HP 346A noise source) and it did a pretty good job across 2MHz to 20MHz, even with the internal HPAK preamp.? This is the only analyser I know of that has a chance of getting good results across this frequency range when using a 346A noise source with its low ENR of about 5dB. I think this particular analyser has an improved preamp in it compared to earlier models. However, the internal preamp only works up to about 3GHz. It offers a noise floor of about -170dBm/Hz down to a few MHz and I think it degrades to about -168dBm/Hz by about 1MHz. This is much better than any other HPAK analyser model that I know of. The input VSWR is quite good too at these frequencies.
?
I haven't used any averaging for either the cal or the measurement and it has given a fairly flat result as expected. I don't actually know what the correct noise figure is for this preamp. The simulation predicted a flat 3.6dB noise figure to below 1MHz but it usually gives a result that is slightly low.
?
I had to tweak my original ENR table at 2MHz to get a flat result at 2MHz though. I reduced it by 0.6dB. I also reduced the 3MHz entry by?about 0.3dB. This is in line with my earlier noise floor measurements using the RSA3408A. This gave a very pleasing result!
?
However, I'll try and tweak the BFR91 preamp for best VSWR and then revisit this with a bit more time. I'll also use averaging and I'll accurately measure the small signal gain with a 2 port VNA (calibrated with an Ecal module).
?
Although this result looks good, I don't think it's worth trying to use it below 2MHz with the 346A noise source as the ENR of this 346A really drops down quite quickly below 2MHz.
?
My homebrew noise source using a Noisecom noise diode is probably more suitable for use below 2MHz as the diode is supposed to work down to 200kHz and HPAK suggest that the analyser can be used down to 200kHz in noise figure mode. I'm not convinced of this though. It will also take ages to do the measurement as it dynamically reduces the detection bandwidth for lower frequencies and this really slows it down.
?
?
|
WReeve