Low frequencies on Spectrum Analyzers

[G0HZU]

Is this normal?  Are there any SpecAns in this price bracket (sub-$2k) that go down lower?  Or is this inherent in this sort of SpecAn design?

There are two big reasons:
1. The first mixer is usually AC coupled through the transformer.  Some spectrum analyzers have an option to swap the mixer inputs for operation down to DC but this risks damage to the mixer if DC is applied.

Most strange... What RF spectrum analysers have an input like this with a transformer and why would you put the DC block at the mixer anyway?

The normal place for the switchable DC block is right at the analyser input ahead of the step attenuator. Often there will be a diode limiter between the attenuator and the mixer. I've not seen a regular lab analyser with a transformer at the mixer input either. I've not looked inside that many analysers but it would be odd to put a transformer at the RF input to the mixer. By transformer I means something with wires wound around a toroid or similar.

[David Hess]

Is this normal?  Are there any SpecAns in this price bracket (sub-$2k) that go down lower?  Or is this inherent in this sort of SpecAn design?

There are two big reasons:
1. The first mixer is usually AC coupled through the transformer.  Some spectrum analyzers have an option to swap the mixer inputs for operation down to DC but this risks damage to the mixer if DC is applied.

Most strange... What RF spectrum analysers have an input like this with a transformer and why would you put the DC block at the mixer anyway?

The normal place for the switchable DC block is right at the analyser input ahead of the step attenuator. Often there will be a diode limiter between the attenuator and the mixer. I've not seen a regular lab analyser with a transformer at the mixer input either. I've not looked inside that many analysers but it would be odd to put a transformer at the RF input to the mixer. By transformer I means something with wires wound around a toroid or similar.

The transformer is at the input in the form of one of the two transformers in the first double balanced mixer.  Below shows an example where the normal IF and RF port on the first mixer are swapped to extend the input range down to DC.  When the diode ring is directly exposed to DC at the input like this, an external AC coupling adapter is recommended to prevent damage because the mixer is awfully easy to burn out in this configuration.

[Performa01]

The transformer is at the input in the form of one of the two transformers in the first double balanced mixer.  Below shows an example where the normal IF and RF port on the first mixer are swapped to extend the input range down to DC.
...

You seem to think this is something special, but it's just the standard arrangement for wideband upconversion receivers (what an SA essentially is) regardless of the lower input frequency limit - and for good reasons.

For an SA to work down to DC the very first and obvious requirement is to have the DC block switchable (or no internal DC block at all).

The performance will be usually poor because upconverting very low input frequencies of a few Hertz into the GHz Range of the 1st IF is equivalent to zooming into the close-in phase noise of the LO.  Any error of just 1ppb in the LO will translate to >10% at 10Hz. Image rejection also gets rather difficult under such conditions.

Because of this, Analyzers that start at DC (or very low frequencies like 10Hz) usually have a dedicated direct processing path with no frequency conversion at all up to e.g. a few hundred kHz.

[G0HZU]

The transformer is at the input in the form of one of the two transformers in the first double balanced mixer.

Ok,  what has thrown me was your use of mixer inputs (plural) as in the quote below. This conjured up an image of a pair of  inputs to a transformer. That would be very strange at this port of the mixer. Usually there is a single feed point. Often this is direct to a pair of diodes and then onto a balun transformer.

The first mixer is usually AC coupled through the transformer.  Some spectrum analyzers have an option to swap the mixer inputs for operation down to DC

Note: I think your block diagram is for a spectrum analyser plug in module for a old Tek oscilloscope? I'm not sure I'd want to try using that at low frequencies for various reasons.

See below for an image of the range 1 mixer for my old HP8566B analyser for example. This doesn't have a DC block anywhere but it does have a diode limiter ahead of this mixer to try and protect it from overload. This analyser works down to about 100Hz although I wouldn't really recommend using it down at AF frequencies. There are so many other options that will be better nowadays.

[David Hess]

The transformer is at the input in the form of one of the two transformers in the first double balanced mixer.  Below shows an example where the normal IF and RF port on the first mixer are swapped to extend the input range down to DC.
...

You seem to think this is something special, but it's just the standard arrangement for wideband upconversion receivers (what an SA essentially is) regardless of the lower input frequency limit - and for good reasons.

For an SA to work down to DC the very first and obvious requirement is to have the DC block switchable (or no internal DC block at all).

It is only special in the sense that spectrum analyzers which wire the first mixer as it is commonly shown cannot operate down to DC and may have a factory option to swap the IF and RF ports.

The performance will be usually poor because upconverting very low input frequencies of a few Hertz into the GHz Range of the 1st IF is equivalent to zooming into the close-in phase noise of the LO.  Any error of just 1ppb in the LO will translate to >10% at 10Hz. Image rejection also gets rather difficult under such conditions.

The local oscillator's phase noise limiting the performance close to DC was my second point.  Did you even read what I posted?

With the phase noise limitation, it makes sense to wire the first mixer with the RF going into one of the transformer ports for better overload protection and I assume that is why it is done.  For applications which require operation down to DC and that can live with low dynamic range, the mixer has to be wired unconventionally and other provisions for overload protection made.  As you point out, a standard spectrum analyser is the wrong instrument for these applications but some people want to do it anyway.

The question about rewiring the first mixer for operation down to DC shows up periodically on various technical email lists.  I get the feeling that most of the old spectrum analyzers with blown first mixers died because of DC applied to the RF input and damaged diodes rather than excessive RF or a damaged transformer.  Maybe newer ones have it but I have never run across one which included a switchable DC block.

The first mixer is usually AC coupled through the transformer.  Some spectrum analyzers have an option to swap the mixer inputs for operation down to DC

Note: I think your block diagram is for a spectrum analyser plug in module for a old Tek oscilloscope? I'm not sure I'd want to try using that at low frequencies for various reasons.

It is but was just the best example I had immediately available for illustrating the point.  Tektronix advertised these for operation down to DC although I am not sure what application they had in mind.

[G0HZU]

Maybe newer ones have it but I have never run across one which included a switchable DC block.

My old Advantest TR4172 has a switchable DC block at the input. This design dates back to about 1981. With the DC block inline it works from 1.8GHz down to 10kHz but with it switched out it says it works down to 50Hz on the front panel. It takes a quick button press to select the DC block. The DC block is right at the input of the analyser ahead of the step attenuator rather than at the mixer itself.

I think my 1.5GHz HP8568B (dates back to the 1970s) works down to 100Hz on one of its inputs. It uses a simple internal RF relay switch to select an input with a DC block or another without a DC block. The default is the N connector with the DC block inline. I rarely used the alternative input without the DC block and left it on the default N connection.

[TAMHAN]

IF (big IF) you can get one cheap, the HP 4195A goes down to 10Hz. I already drove it there, and it worked quite well.

Of course, it is a heterodyne sweep unit, and is not the fastest to sweep.

Tam

[G0HZU]

We used to have an HP4195A at work. It's a very versatile instrument. However, it is very slow to sweep. For a bit of fun at work I think we managed to get ours to display a sweep time of 25 years by setting it to some silly settings. I recall it actually gave the sweep time in years but I can't be certain. It was about 25 years ago so maybe it would have finished by now if we had left it alone all this time 

[Performa01]

We used to have an HP4195A at work. It's a very versatile instrument. However, it is very slow to sweep. For a bit of fun at work I think we managed to get ours to display a sweep time of 25 years by setting it to some silly settings. I recall it actually gave the sweep time in years but I can't be certain. It was about 25 years ago so maybe it would have finished by now if we had left it alone all this time 

Pity you didn't do that. Now we'll never know how accurate the sweep time estimation of this instrument has been!

[G0HZU]

I miss having the HP4195A at work because it worked so well as a VNA or a spectrum analyser at low frequencies. But it was costing too much in yearly calibration to justify us keeping it. It often went unused for a year or more so it got disposed of.

My 1.5GHz HP8568B doesn't really suffer from the LO noise limitation at low frequencies. It typically manages a noise floor of about -149dBm/Hz up at VHF. It only degrades by about 6dB or so down at audio frequencies and this is with the correct 100Hz+ input selected. So it can show about -143dBm/Hz typical noise floor across a span of 0-10kHz.

Other analysers will usually show a steep rise in noise floor below about 200kHz but my HP8568B is very good here. Much better than the datasheet limits. But I still don't use it down at these frequencies unless I'm looking for very tiny signals and can afford to wait for the slow sweep time.

[Performa01]

Do you happen to know how the 8568B manages to provide such a decent low frequency performance?
Separate band with much lower 1st IF?

I don't know this oldie, but I guess it's a safe bet that it doesn't have any means for digital signal processing?

[G0HZU]

It was an extremely expensive analyser that was designed to try and maintain as much performance as possible over its full frequency range. It has a synthesised LO scheme with a decent internal 10MHz OCXO and also has very low close to carrier phase noise on the first LO and so this also means it has an advantage over other analysers here. So my HP8568B can achieve maybe -143dBm/Hz noise floor even down at 1kHz. However, there will probably be quite a spread in this level across various examples of this analyser. This level of performance is much better than the 'limit' figures in the datasheet and maybe not all HP8568B analysers will be as good as mine. Generally, the typical performance for the HP8568B analyser is a lot better than the datasheet limits in a lot of areas.

It achieves this performance without needing the separate direct path to the IF. The AF signals get upconverted to a 2GHz IF just like all other signals.

It can do some DSP tricks (eg post detection FFT with a choice of the classic Hanning or Flat top windowing) but not in the regular analyser mode.

[G0HZU]

Here is a plot of my HP8568B with a span of 0-2kHz with a RBW of 30Hz. It can go down to 10Hz RBW but it would takes ages to sweep.
Sorry for the gloomy image as my camera doesn't take pictures of CRTs very well and also the CRT is getting a bit dim in this analyser. It hasn't been used for ages and I think the CRT might benefit from me leaving it on for a while. Because the noise level is so low, some people might think the analyser is faulty in some way but I can send a known test signal to it and it shows up correctly in terms of level. This was a very special instrument in its day. The display is digitised even though it is a CRT and it has 1000x1000 datapoints in the grid area. This is still good even today.

[Performa01]

I love challenges!

Anyway, here comes my answer to the question what single device can “do both”, i.e. works for very low as well as high frequencies. When I saw the noise floor of the ancient HP8568B, I thought “oh boy, this would be hard to beat” – but I decided to give it a try anyway.

My contender is the Signal Hound SA44, which works in the range of 1Hz to 4.4GHz (Tracking generator TG44 from 10Hz to 4.4GHz).

Now the question is … is it fair to place a modern low cost SA against an ancient high end boat anchor (the datasheet says 45kg!) that has probably cost its first owner an arm and both legs?

The answer can only be a resounding “Yes”!

Both are affordable options today, even though for the HP8568B one has to calculate the additional costs for a crane truck as well as another (reinforced) workbench – but I guess these are minor considerations for a true test gear aficionado…

First I tried to resemble the scene shown by G0HZU as close as possible, just had to use 25Hz RBW because 30Hz isn’t available on a SA44. No input signal, I’ve just terminated the input with 50 ohms.


SA44_DANL_2kHz_F

Voilà, David has scored against Goliath. A DANL of -149dBm/Hz should certainly be good enough for the majority of practical applications in this frequency range.

However, a low noise floor alone does not make us happy – we are more interested in the signal to noise ratio, and for that we need a signal. When I tried this, I immediately realized why G0HZU preferred to show the noise floor without a signal: at frequencies that low, we can get all kinds of interferences by connecting the cable alone, even when the signal source is still off. In fact I’ve verified that the ground connection to any device (that isn’t even powered on) quite obviously causes a ground loop with lots of unwanted signals related to mains hum and its harmonics at levels up to -110dBm below 1kHz, so I had to measure the noise floor at 1500Hz this time:


SA44_DANL_2kHz_1000Hz_-40dBm_F

Now I would love to see how Goliath behaves in this test scenario, and I have the sneaky feeling this could already be a decisive battle in this contest…

The noise floor is a bit higher now by about 4dB and without more tests we cannot be absolutely sure where it comes from – the signal generator, the ground loop or the analyzer itself when it sees a strong signal.

So here’s the exact same measurement with the signal generator output off and powered off on top of that. As can be seen, the noise floor and the interferences are pretty much the same, so it’s all coming from the ground loop:


SA44_DANL_2kHz_GroundLoop_F

In any case the signal to noise ratio is close to 90dB in this scenario, but for sensitive measurements one would really have to find a cure for the mains hum and other interferences caused by ground loops. Shielded audio transformers might be an option, but are usually not made for 50 ohms and the frequency response might become problematic – apart from the fact that this obviously cannot work down to DC anyway.

Now finally I need to prove that this analyzer can not only work at very low frequencies, but also higher ones. Since the HP8568B works up to 1.5GHz, I’ve made another measurement at that frequency. Of course I’ve taken advantage on the built-in preamplifier, that can be engaged for frequencies above 100kHz and that’s also the reason why the noise is now even lower at -164dBm/Hz:


SA44_DANL_2kHz_1500MHz_-100dBm


WARNING: The SA44 does not have an internal DC block – of course not. During normal use, I always have an external one attached to its input and one really needs to be careful when working without. The input cannot tolerate DC at all and any voltage exceeding +/-100mV might damage it permanently.

[G0HZU]

I think your SH44 gives good performance at low frequencies because it uses a very low first IF of something like 10.7MHz. So the first LO runs at just a few MHz and it's easy to get low phase noise and good mixer balance at a few MHz. I'm going to guess your SH44 uses something like an ADF4350 for the first LO and this can get very good phase noise when the GHz VCO in the ADF4350 is divided down inside the chip to just a few MHz.

For the designers of the HP8568 back in the mid 1970s they had to do this with an IF of 2GHz. Much harder because the LO runs at >2GHz! Normally, you can kind of predict the very close to carrier phase noise of a well designed LO at 2GHz with a few guesstimates and a simple sum.

If we assume the 10MHz reference OCXO inside the analyser is a good one with maybe -150dBc/Hz phase noise at 500Hz offset then the phase noise at 500Hz offset at the 2GHz first LO could normally be as good as -150 + 20*log(2000/10) = -104dBc/Hz.

But the HP8568B manages about -112dBc/Hz here and it does it with a clever comb generation scheme in the first LO which is able to dodge the 20*logN part of the equation at lowish analyser frequencies. i.e. at analyser centre frequencies up to maybe 200MHz or so. This means the HP8568B phase noise is very 'flat' at -112dBc/Hz from carrier offsets of a few 100Hz to maybe 80kHz offset across analyser frequencies of LF/HF/VHF. In order to see a -143dBm/Hz noise floor at a 500Hz input frequency the first mixer will need to have good balance to prevent bleedthrough of the first LO.

So I assume that the mixer must have about (143-112) + 13 =  44dB isolation to the mixer IF port if the first LO runs at maybe +13dBm. The conversion loss of the mixer (and the front end attenuation) should also be factored in but this -143dBm/Hz noise floor at sub 1kHz is an impressive result with an LO running at 2GHz.

I immediately realized why G0HZU preferred to show the noise floor without a signal

My HP8568B is stored downstairs so I can't easily connect any decent test gear to it. I don't use it much anymore. I just unpacked it and turned it on and took a picture. I did try testing it yesterday at 10kHz with a little Marconi 2022 sig gen at a known -80dBm to prove the analyser was healthy. The little 2022 generator was easy to carry downstairs but it can't go below 10kHz. But I've been using this same HP8568B analyser since 1990 at work and now here at home and I know its performance well.

[G0HZU]

Now finally I need to prove that this analyzer can not only work at very low frequencies, but also higher ones. Since the HP8568B works up to 1.5GHz, I’ve made another measurement at that frequency. Of course I’ve taken advantage on the built-in preamplifier, that can be engaged for frequencies above 100kHz and that’s also the reason why the noise is now even lower at -164dBm/Hz:

It's easy to get a low noise floor if the analyser has a preamplifier built in. My old Advantest TR4172 has an internal switchable preamp but I rarely use the preamp as it compromises the analyser performance in too many ways.
The preamp has a lot of gain and I think the noise floor is typically -166dBm/Hz with the preamp in but I've never bothered to really investigate the preamp

[rf-loop]

Just for... (image is self explanating - no comments)

[Performa01]

@G0HZU

I do hope you didn't get me wrong in any way. I fully respect the HP8568B and its remarkable performance with regard to the architecture used. And because I usually don't use the SA44 down below some 10kHz (which requires to remove the DC block) I had no idea if it was able to keep up or even beat the HP.

Of course the architecture of the SA44 is vastly different, that's the reason why I thought it might be able to compete despite the low cost. Unsurprisingly, the concept has its limitations, but still works surprisingly well for many applications, especially narrowband ones. And at the end of the day most users  are interested in the results only...

I also hope you can forgive what I said about you not showing a signal - do not take it too seriously please

With regard to the preamplifier, you seem to be one of those purists who say that a good receiver frontend has to have a passive high level double balanced mixer, with nothing but a switchable attenuator and a pre-selector in front of it, but by no means a preamplifier. If so, then we're not far apart

Yet in the rare cases where we deal with very low signals exclusively, what's wrong with the preamp then?

In case of the SA44, I have only once measured it at 30MHz. The CP changes from +0.5dBm to some -18dBm with the preamp on.

I've just measured the DANL at 1.5GHz again. Without preamp it is about -151dBm/Hz - marginally better than at 1kHz, where no preamp is available from the outset.

[Performa01]

Just for... (image is self explanating - no comments)

Nice! 

Guess I'll need to have a look into that eventually.

Yet I think you're cheating a bit ... might it be that the input signal is actually a bit higher than -40dBm?

[Performa01]

Just in case any Goliath is still in combat mode, here are a few more low frequency tests taking advantage on the narrow resolution bandwidths of the SA44 down to 0.1Hz.

First let’s have a look at the DANL at really low frequencies from 0-60Hz at 0.1Hz RBW. We can see the mains signal at 50Hz together with two spurs at 40 and 60Hz that look like sidebands from some 10Hz modulation. The Noise level is significantly higher down here, yet -142dBm/Hz should not cause any troubles in real applications.


SA44_DANL_60Hz_MF

Below are a couple of experiments that take advantage of the high frequency resolution. Sweep is slow at about 37.5s, but that’s just the price for very narrow resolution bandwidths according to the laws of physics (and digital signal processing).

We start with a 20Hz carrier, 100% amplitude modulated at 1Hz. Regular markers have been placed on the carrier, the upper sideband and the mains hum. There are many high order harmonics of the modulation frequency visible, but at relatively low levels.


SA44_DANL_60Hz_20Hz_AM100%_1Hz_MF


The next screenshot shows a 3Hz square wave at -41dBm. All the odd harmonics are clearly visible, whereas the even harmonics are missing, just as it has been written in the textbooks. The 50Hz mains hum is clearly distinguishable from the 17th harmonic at 51Hz.


SA44_60Hz_Square_3Hz_-41dBm_MF


I happened to catch the moment when I changed the signal level from -40dBm to -41dBm. For this amplitude step some internal relay in the generator was switching and the SA44 captured some interesting looking piece of art:


SA44_60Hz_Square_3Hz_-40dBm_SW_MF


Finally I wanted to create a comb pattern where all harmonics are equal in amplitude – at least up to 60Hz. I used a 1ms wide pulse with 2ns rise & fall times at a repetition rate of 3Hz for that. The pulse amplitude is -40dBm again, but its energy is now spread over a wide frequency range, hence the measured level is only -80dBm.


SA44_60Hz_Pulse_1ms_3Hz_-40dBm_MF

[G0HZU]

Pretty good from the Siglent It is within about 10dB of my HP8568B in terms of noise level and that is a lot better than a lot of expensive analysers we have at work from R&S. Because the Siglent uses FFT it's also going to sweep a lot faster than my old HP8568B with its old school analogue RBW filters.

Here's a quick video of my 8GHz Tek RSA3408A when used down at 0-5kHz. This shows the noise and harmonics from an old audio generator I made when I was a student. It uses cheapo opamps so the noise and distortion isn't that great. But the Tek RSA3408A is quite fast in terms of update rate. However, this is still classed as an old and obsolete analyser at work. I got this one for free, saved from the dumpster at work because it had a minor fault. Now working

https://www.youtube.com/watch?v=G7SSoe3Bils&feature=youtu.be

[Performa01]

Here's a quick video of my 8GHz Tek RSA3408A when used down at 0-5kHz.

Pity I cannot view it now with my old PC because of the HTML5 format...

[G0HZU]

I tried uploading again here:
https://www.youtube.com/watch?v=G-OkogKFxKg&feature=youtu.be

Does this one play for you or can you not watch any youtube videos on your PC?

[Performa01]

I usually can - but not these modern content protected (I guess) video formats - I'm still on Win XP here in the lab...

EDIT: Guess it's the H.264 fromat that is not supported.

[G0HZU]

Try again here:

https://www.youtube.com/watch?v=uqthKonp9Xc&feature=youtu.be


or this one

https://www.youtube.com/watch?v=GqttvZw76_g&feature=youtu.be

The second one is really grotty with a stretched aspect ratio but it is the most basic format I can find?