My day job is in RF, so when I want to see a signal a spec-an is my go-to instrument. Way back when I was building some amps, I dug around the lab at work to find a spec-an that would be useful for audio, but came up well short. The microwave instruments (Rohde & Schwarz ZVL etc) typically only go town to 9KHz or so, and don't have very good dynamic range anyway.
Anyway, recently I got my hands on a truly ancient HP3585B. It was dead as a doornail, and it was going to the tip. It had a full set of service manuals for it, so I thought I'd have a squiz and see if I couldn't breathe some life into it. My workmates scoffed because it has a maximum frequency of 40MHz, rather than the more typical 20 odd GHz of instruments we use, plus it weighs about 40kg. It's really seriously big and heavy, so much so that I need to get help to move it around. Anyway, I was more fixated on the 20Hz minimum frequency than the 40MHz maximum...
It came with a friend - a HP3325B 20 MHz function generator (similarly bulky but thankfully nowhere near as heavy, similar vintage, also quite dead).
After filling the lab at work with smoke on attempting to power up each instrument, they were summarily evicted and came home to live.
The problems with both were all power supply - a cap in the input mains filter on the 3325B had gone (and let out a tremendous amount of smoke), and a tantalum on the CPU board had similarly let go on the 3585B. The spectrum analyser had more wrong than that though - I wasn't getting any EHT so had no display on the screen. I eventually traced that to a dead diode in the EHT supply, sourced a replacement, put it in and voila!
So the first measurement: generate a 1KHz tone from the 3325B and measure that with the 3585B:
Yeah, nothing to write home about. That's what, 70, 75dB from carrier to second harmonic...
At that both went into the spare room and languished for a while.
Anyway, this weekend I pulled the spec-an out to look at an amplifier, as I wanted to see some oscillations at around 1.8MHz. On a whim, I connected it to my mac (a 2011 MacBook Air), and used Audacity to generate a 1KHz tone (96KHz sample rate, 24 bit samples):
Now we're cooking with gas. I reckon that's gotta be 95dB SNR.
Zooming in a little shows all the distortion, in the second and third harmonic, plus well-suppressed 50 and 150Hz:
Anyway, I reckon that's an instrument that I can learn to love 🙂
Anyway, recently I got my hands on a truly ancient HP3585B. It was dead as a doornail, and it was going to the tip. It had a full set of service manuals for it, so I thought I'd have a squiz and see if I couldn't breathe some life into it. My workmates scoffed because it has a maximum frequency of 40MHz, rather than the more typical 20 odd GHz of instruments we use, plus it weighs about 40kg. It's really seriously big and heavy, so much so that I need to get help to move it around. Anyway, I was more fixated on the 20Hz minimum frequency than the 40MHz maximum...
It came with a friend - a HP3325B 20 MHz function generator (similarly bulky but thankfully nowhere near as heavy, similar vintage, also quite dead).
After filling the lab at work with smoke on attempting to power up each instrument, they were summarily evicted and came home to live.
The problems with both were all power supply - a cap in the input mains filter on the 3325B had gone (and let out a tremendous amount of smoke), and a tantalum on the CPU board had similarly let go on the 3585B. The spectrum analyser had more wrong than that though - I wasn't getting any EHT so had no display on the screen. I eventually traced that to a dead diode in the EHT supply, sourced a replacement, put it in and voila!
So the first measurement: generate a 1KHz tone from the 3325B and measure that with the 3585B:
Yeah, nothing to write home about. That's what, 70, 75dB from carrier to second harmonic...
At that both went into the spare room and languished for a while.
Anyway, this weekend I pulled the spec-an out to look at an amplifier, as I wanted to see some oscillations at around 1.8MHz. On a whim, I connected it to my mac (a 2011 MacBook Air), and used Audacity to generate a 1KHz tone (96KHz sample rate, 24 bit samples):
Now we're cooking with gas. I reckon that's gotta be 95dB SNR.
Zooming in a little shows all the distortion, in the second and third harmonic, plus well-suppressed 50 and 150Hz:
Anyway, I reckon that's an instrument that I can learn to love 🙂
Nice. The 95dB range is nothing to scoff at, and if all your audio stuff gets down to this level you are doing quite well!
Jan
Jan
I wonder what a clean-up of front end can do to improve performance by 10-20dB? Like better opamps on input - if it used them. Or what?
THx-RNMarsh
THx-RNMarsh
One of the best pieces of kit ever, sat on my bench for years. It got dim but I put up with it, someone else sent it to the cal lab and they tossed it because the CRT replacement cost an arm and a leg. Computed nV/rt-Hz directly, 7nV noise floor.
Dick I think this is a triple conversion RF instrument no op-amps to roll. Yup, 3Hz wide crystal ladder filters IIRC. I don't think anyone has an RF based instrument that does better than -100db spuriae.
Last edited:
I used a 3325 in a previous job and love the precision of it.
I have access to a 3324 (non-50R version) and did not like the "<1%" THD spec. 🙁
Still a nice generator.
I have access to a 3324 (non-50R version) and did not like the "<1%" THD spec. 🙁
Still a nice generator.
I've not played with its measurement abilities, other than to throw markers up. I shall investigate. I thought similarly that the CRT was dead on this one, and was quite pleased when I found there was simply no EHT, as I knew that would be a much easier fix (though it turned out to be a challenge to measure, as I haven't had any HV probes in years, so jury-rigged one from a stack of 1M resistors).
It is indeed a purely RF instrument. No fast ADCs, just conversions and gorgeous narrow filters. That means the noise floor is just as low measuring a 10MHz tone as a 1KHz one. I got the full 100dB measuring a 10MHz OCXO. Indeed both the 3325 and 3585 came with low-noise references, which is nice.
I've been working the last few days in really using this instrument in anger on one of my audio power amps. Firstly, the noise density readout is incredibly cool. Here's a plot of my amp output, with noise calculated at 1KHz:
That equates to 8.7 nV/sqrtHz input referred (gain of 16) at 1KHz.
Getting really good THD measurements has proven to be a real challenge. I started by just throwing a 20dB pad between my amp output and the HP3585, but hit the spec-an DNR limitation straight away. Then I built a 1KHz passive twin-T notch, which helped quite a lot. I used the tracking generator to tweak the notch to give me good rejection at 1KHz, and to measure attenuation at H2, H3, etc. The tracking generator is also useful for checking plotting the amp gain curve:
It's a measurement that I'm still refining, but here's what I've got so far:
Gotta say, I'm stoked at that. Best I was able to measure previously was 0.0011%, using a borrowed Tek AA501. Being able to push measurements properly under 0.001% is way cool.
The technique is super-tedious. Start by measuring the fundamental with the 20dB pad to protect the analyser input, then substitute the notch, zero in on H2 (0 Hz span sets 3 Hz RBW), wind the range down until the clipping indicator almost comes on, then wind the reference level down until the trace comes off the bottom of the screen.
Write down the number and repeat for all H2-H10 at all amplitude levels, then throw the whole mess into a spreadsheet to calculate THD.
Note the MacBook alone (which I was using as a source for my amp) measures worse than with the amp connected for lower power levels. This is just a noise floor issue. I really need a 20dB zero-noise, zero-distortion preamp 🙂 I'm wondering if the closest thing I've got to that is just another of my power amps strapped to gain of 10, with no load resistor.
That equates to 8.7 nV/sqrtHz input referred (gain of 16) at 1KHz.
Getting really good THD measurements has proven to be a real challenge. I started by just throwing a 20dB pad between my amp output and the HP3585, but hit the spec-an DNR limitation straight away. Then I built a 1KHz passive twin-T notch, which helped quite a lot. I used the tracking generator to tweak the notch to give me good rejection at 1KHz, and to measure attenuation at H2, H3, etc. The tracking generator is also useful for checking plotting the amp gain curve:
It's a measurement that I'm still refining, but here's what I've got so far:
Gotta say, I'm stoked at that. Best I was able to measure previously was 0.0011%, using a borrowed Tek AA501. Being able to push measurements properly under 0.001% is way cool.
The technique is super-tedious. Start by measuring the fundamental with the 20dB pad to protect the analyser input, then substitute the notch, zero in on H2 (0 Hz span sets 3 Hz RBW), wind the range down until the clipping indicator almost comes on, then wind the reference level down until the trace comes off the bottom of the screen.
Write down the number and repeat for all H2-H10 at all amplitude levels, then throw the whole mess into a spreadsheet to calculate THD.
Note the MacBook alone (which I was using as a source for my amp) measures worse than with the amp connected for lower power levels. This is just a noise floor issue. I really need a 20dB zero-noise, zero-distortion preamp 🙂 I'm wondering if the closest thing I've got to that is just another of my power amps strapped to gain of 10, with no load resistor.
Attachments
Last edited by a moderator:
I've been getting a lot of "LO unlocked" errors on my HP3585B.
I've ignored them studiously until now, on the basis that if I go away and make a cup of tea, the instrument generally gets it's act together by the time I get back.
Anyway, they got on my wick enough to force me to troubleshoot the problem.
Most of the timing is derived from a 90MHz crystal in the spec-an. This is either free-running, or else (if you've got the high-stability reference version, as I have) it's phase locked to a reasonable quality OCXO.
So I calibrated a spare 5MHz Wenzel OCXO at work to a reasonably accurate 5MHz source, and took that home with me to use as a reference. I did this on the basis that the 3585 is too bloody heavy to drag in to work with me, and the standards at work are _way_ too big and heavy to take home (not to mention that they'd be missed if I borrowed one overnight).
So I figured I could live with the millihertz level of uncertainty that I'd get by setting up an OCXO and then moving it.
Anyway, plugging that into the spec an showed it as 5,000,002.6Hz, so the OCXO in the spec-an was pretty good. I tried tweaking the spec-an reference oscillator and straight away got the "LO unlocked" message, and my 5MHz tone was now about 150Hz out.
Wondering if there might be something wrong with the spec-an OCXO I unplugged it at the back - my tone stayed about 150Hz out. I checked the 10MHz out from the spec-an with my CRO and it looked okay - or at least as okay as a CRO is going to tell you.
So figuring I was now running on the internal non-stabilised timebase, I figured I should tweak that to get it closer to correct, so I adjusted the 90MHz oscillator to pull my 5MHz reference back to 5MHz, or at least within a couple Hz - the adjustment just wasn't really up to finessing.
I plugged the main reference back in and now it was solid, I finished tweaking that and no more LO unlocked errors.
I guess the error between the 90MHz oscillator and the reference was just too big for the machine to pull in by itself.
Anyway - it's happy now - touch wood that I don't have to pull it apart again, as it's a bloody heavy lump of metal!
I've ignored them studiously until now, on the basis that if I go away and make a cup of tea, the instrument generally gets it's act together by the time I get back.
Anyway, they got on my wick enough to force me to troubleshoot the problem.
Most of the timing is derived from a 90MHz crystal in the spec-an. This is either free-running, or else (if you've got the high-stability reference version, as I have) it's phase locked to a reasonable quality OCXO.
So I calibrated a spare 5MHz Wenzel OCXO at work to a reasonably accurate 5MHz source, and took that home with me to use as a reference. I did this on the basis that the 3585 is too bloody heavy to drag in to work with me, and the standards at work are _way_ too big and heavy to take home (not to mention that they'd be missed if I borrowed one overnight).
So I figured I could live with the millihertz level of uncertainty that I'd get by setting up an OCXO and then moving it.
Anyway, plugging that into the spec an showed it as 5,000,002.6Hz, so the OCXO in the spec-an was pretty good. I tried tweaking the spec-an reference oscillator and straight away got the "LO unlocked" message, and my 5MHz tone was now about 150Hz out.
Wondering if there might be something wrong with the spec-an OCXO I unplugged it at the back - my tone stayed about 150Hz out. I checked the 10MHz out from the spec-an with my CRO and it looked okay - or at least as okay as a CRO is going to tell you.
So figuring I was now running on the internal non-stabilised timebase, I figured I should tweak that to get it closer to correct, so I adjusted the 90MHz oscillator to pull my 5MHz reference back to 5MHz, or at least within a couple Hz - the adjustment just wasn't really up to finessing.
I plugged the main reference back in and now it was solid, I finished tweaking that and no more LO unlocked errors.
I guess the error between the 90MHz oscillator and the reference was just too big for the machine to pull in by itself.
Anyway - it's happy now - touch wood that I don't have to pull it apart again, as it's a bloody heavy lump of metal!
I thought I'd work on getting proper plots out of my 3585B.
It only has a GPIB port, so I bought a GPIB-USB cable. Being a tight-****, I found an NI GPIB-USB-A for cheap on the interwebs, so bought that.
So then I found out why the NI GPIB-USB-A cable was cheap - turns out no modern drivers support it, as it's a few versions old. I figured this was a challenge, and don't particularly care if I'm using the latest versions of things.
So after some google-foo, I found I need version 2.2 of NI-488.2, which only works on Windows 98, Windows Me, Windows 2000 and Windows XP. NI hide their old driver software _very_ well. After much digging I found it on their ftp server, at ftp://ftp.ni.com/support/gpib/
Okay. So now I bought a PC. It's been a decade since I've owned a Windows box - normally I use macs. Again, being a tight-****, I bought a super-cheap old HP (hey, HP PC for talking to a HP instrument, right?) DC7900 small form factor one, as I didn't want to waste any more desk space than I absolutely needed. The I bought an old XP install DVD, some RAM, and an SSD - my cheap PC is now costing a couple of hundred bucks. Finally tonight I got windows XP (rather than windows 7) running on it, and installed the NI 488.2 drivers.
All good. I can go into my "instrument explorer" and see my CRO, my synthesiser, my spec-an and my multimeter all at once. Yay.
So finally I installed a free HPGL plotter emulator, from 7470: HP-GL/2 Plotter Emulator
I run it, put it in listen mode, press the plot key on my 3585B, and get a real honest for god plot. It's like I've joined the 20th century. No more messing around with taking photos of the screen 🙂
It only has a GPIB port, so I bought a GPIB-USB cable. Being a tight-****, I found an NI GPIB-USB-A for cheap on the interwebs, so bought that.
So then I found out why the NI GPIB-USB-A cable was cheap - turns out no modern drivers support it, as it's a few versions old. I figured this was a challenge, and don't particularly care if I'm using the latest versions of things.
So after some google-foo, I found I need version 2.2 of NI-488.2, which only works on Windows 98, Windows Me, Windows 2000 and Windows XP. NI hide their old driver software _very_ well. After much digging I found it on their ftp server, at ftp://ftp.ni.com/support/gpib/
Okay. So now I bought a PC. It's been a decade since I've owned a Windows box - normally I use macs. Again, being a tight-****, I bought a super-cheap old HP (hey, HP PC for talking to a HP instrument, right?) DC7900 small form factor one, as I didn't want to waste any more desk space than I absolutely needed. The I bought an old XP install DVD, some RAM, and an SSD - my cheap PC is now costing a couple of hundred bucks. Finally tonight I got windows XP (rather than windows 7) running on it, and installed the NI 488.2 drivers.
All good. I can go into my "instrument explorer" and see my CRO, my synthesiser, my spec-an and my multimeter all at once. Yay.
So finally I installed a free HPGL plotter emulator, from 7470: HP-GL/2 Plotter Emulator
I run it, put it in listen mode, press the plot key on my 3585B, and get a real honest for god plot. It's like I've joined the 20th century. No more messing around with taking photos of the screen 🙂
Attachments
Last edited:
- Status
- Not open for further replies.