Low frequencies on Spectrum Analyzers

[Tomorokoshi]

This just showed up on ebay:
https://www.ebay.com/itm/Agilent-HP-35665A-DSA-Dynamic-Signal-Analyzer-w-Opt-1D2-1D4-1C2-ANA/142741758701

These go to around 244 micro Hz in the right conditions.

It's something of an update to these:

Most of the "RF" spectrum analyzers have to worry about local oscillator feedthrough and can't get too close to DC.  If you are willing to go old, the
HP 3561A or HP 3562A are very nice 0-100 kHz analyzers.

Or in a pinch, I've always found that using the sound card of a computer and a freeware FFT program gives very reasonable results.

And relatively similar to this:

You want an SR780!

Dave has a video on the machine that is slightly older:
https://www.eevblog.com/2013/09/21/eevblog-523-repair-hp-35660a-dynamic-signal-analyser/

An eeprom upgrade to the 35665A to enable all software features is available at:
http://www.glkinst.com/test-equipment/

[Performa01]

@G0HZU Thank you very much - both recordings play fine on my ancient machine.

And about the contents - looks impressive indeed, especially for an 8GHz SA. There is not much increase of the noise floor when the signal is applied. Makes me wonder what the 1st IF in this scenario would be...

[G0HZU]

The RSA3408A is straight to the ADC below 40MHz so it has very good performance here.

For completeness I took my little AF generator downstairs and tried it on the HP8568B. See below for a 2kHz span as before but this time with the signal applied and I've used a 10Hz RBW. You can see the noise is higher and this is from my generator rather than any reciprocal effects from the analyser phase noise. Sorry the picture is a bit fuzzy, the camera doesn't focus well on CRTs.

[Performa01]

Oh yes, direct to the ADC is certainly the method of choice for a good modern analyzer - and it certainly pays off. 40MHz is quite high - do you happen to know how many bits the ADC has?

The screenshot is plenty good enough and the performance appears actually very good. All the more so considering the signal source is just an RC audio generator using standard op-amps. 

[G0HZU]

I think the ADC is 14 bit and this one has the extended memory option when in real time mode. It's at its best when capturing wideband RF signals (up to 36MHz capture BW) for post processing and that's what we use them for at work. However, we generally need more real time BW than 36MHz these days so they are slowly being replaced with modern stuff from Keysight with three times the BW.

The RSA3408A user interface is classic Tektronix, poorly organised and laggy and it has some really bizarre bugs when used near its limits. However, I'm not that impressed with the Keysight replacements (PXA and EXA) in terms of the user interface. The latest versions of these analysers use a touch screen interface and you need tiny fingers and the presentation of the touch menus is poor.

[Tomorokoshi]

Here is a shot of the HP 35665 showing an FFT on two channels. It's showing the output of an HP 203A dual-output signal generator.

The 203A was supposed to be outputting 100 mHz, but it's actually around 115 mHz. It was okay at higher frequencies. Another repair project.

It takes between 30 minutes and 60 minutes at the resolution and channel setup to get that trace. The timer was around 3600 seconds, but it seemed to be updating something at twice that rate. That could be figured out more easily at a higher frequency.

[G0HZU]

Impressive stuff! I'm not sure what the RSA3408A can do 'directly' with a 0.1Hz test signal but I can easily set it up to grab I and Q data of a 0.1Hz signal. I've left it on a 5kHz span so probably sampling at something like 6.4ksps. But the idea below is just to quickly show it can at least recognise that a 0.1Hz signal is being fed to it.

See below for a 20 second span taken from a 50 second data grab of a 0.1Hz test signal at this sample rate. Because it's a Windows machine it should be possible to write some code to post process and display this data in the frequency domain.

Ultimately, I think I can turn down the sample rate to something like 128sps and grab up to 256M of data for post processing. But I think that would take over 100 hours to complete!

[Performa01]

I’m not entirely sure if frequencies below 1Hz still qualify as just ‘low’

Anyway, the SA44 is not useful below 1Hz anymore. Not because the datasheet says so, but the lowest resolution bandwidth is 0.1Hz and quite obviously spectrum analysis becomes problematic when the signal frequency approaches the resolution bandwidth.

For challenges like this, I had to pull another weapon, it’s the PicoScope 4262. Once again a little David that isn’t afraid to match with the various Goliath boat anchors out there

The lowest analysis bandwidth is 100Hz, but we can zoom in as much as we like (x100 in this example) and take a closer look at the 0-1Hz segment.

Just to make sure that David cannot lose, the RBW has been set to some 725uHz, resulting from a 524284 point FFT thus 381.5uHz frequency step and 725uHz RBW with the Blackman-Harris window.

Dual channel FFT? Well, why not?!

•   Channel A : 1Vrms 100mHz sine wave
•   Channel B : 1Vrms 200mHz sine wave

One sweep takes close to 44 minutes, so should be in the same ballpark as the HP35665.


Pico 4262 FFT-1Hz_2

Harmonic distortion of the input signals is below -70dB and all the harmonics are clearly visible.
Note the dynamic range (113dB) and SNR (110dB) in this measurement.

Also keep in mind that this is just 1% of the whole picture. We can adapt the frequency window by changing position and zoom factor anytime we like. The screenshot below uses the same zoom factor, but different frequency areas. The 100mHz signal for instance has still some odd harmonics even above 27Hz. For the 200mHz signal, the range 1-2Hz is shown, where we can see the last harmonic at 1.6Hz.


Pico 4262 FFT-1Hz_3

[G0HZU]

I had a go with the RSA3408A today and this was my first attempt at a 2Hz span. I've had to post process the data in the 89600 SW but this SW could run on the 3408A in the background.

Note that the full span is 100Hz and the 0.1Hz signal show distortion up to about 20Hz where there is the last distortion term visible at -98dBc. After that there is nothing in the way of distortion all the way up to 100Hz. There is a single wobbly ~50Hz term that is not quite 50Hz at -77dBc but this is going to be mains related ground leakage I think.

But the plot below shows the 100mHz signal and all the harmonics to 2Hz. I went for a 1mHz RBW to keep my sample rate vs sample time numbers simple but it can go much lower than this. I suspect it could do 0.1mHz RBW (maybe even 0.01mHz) if I took enough samples as I think the buffer in the 3408A is good for >100 hours of data at this sample rate. But this is already very silly. I wonder how well this would work with just an Arduino and a decent ADC at a low sample rate

[thanasisk]

The attenuation/level matching problem when using a soundcard may be solved using solution like this:

https://linearaudio.nl/la-autoranger

Or a 20dB pad/attenuator (or even a x10 scope probe)  may work well forallowing a reasonable range of input amplitudes.

[nctnico]

The attenuation/level matching problem when using a soundcard may be solved using solution like this:

https://linearaudio.nl/la-autoranger

What does it do? It doesn't say that on the website.

[ocw]

Attached is a look at 0.1 Hz on my Stanford Research SR760.  My generator was set at 100 mHz while the SR760 measured that as 100.136 mHz. 
There is the inconvenience of 1048.6 seconds (about 17.5 minutes) of acquisition time for that low of a frequency span.  And the SR760's span is 100 kHz or that divided by a power of two.  So, I had it set at 100,000 Hz / 262,144 = 381.470 mHz.

[ocw]

I took my SR760 down to its minimum 191 mHz span for viewing a 30 mHz signal as shown on the attachment.

[thanasisk]

The attenuation/level matching problem when using a soundcard may be solved using solution like this:

https://linearaudio.nl/la-autoranger

What does it do? It doesn't say that on the website.

It is an automatic attenuator/agc,  i think based on rms measurements of the input signal.  It makes sure that the output is at a selected level (e.g. 1 volt) and absolute maximum of 5 volts to protect the sound card

The info is spread here and there and there is a relevant forum http://www.diyaudio.com/forums/equipment-and-tools/299635-autoranger-soundcards.html

[radiolistener]

I think it's better to use good sound card for low frequency. It will provide good dynamic range, due to 24 bit ADC...