Great disassembly job. There's yet another branch of the rabbit hole that Tom almost fell down:
... the multiplier is based on a classic form,
having a translinear core, supported by three
(X, Y, and Z) linearized voltage-to-current
converters, and the load driving output amplifier.
I have no clue how the thing works!
The 'translinear' school of design was founded (or at least popularized) by Barrie Gilbert at Analog Devices (https://en.wikipedia.org/wiki/Translinear_circuit , https://en.wikipedia.org/wiki/Barrie_Gilbert ). Gilbert was one of the undisputed GOATs in the analog field and a full-fledged Renaissance man in his own right. It's worth reading more about him if/when you have the time and inclination.
It would be interesting to benchmark this against consumer USB DACs that are marketed for sound. You can get 32-bit 192kHz bandwidth (384kHz sampling rate) DACs now for under $200. Obviously this is much lower bandwidth than the R&S, but I'd be interested to see the dynamic range of a modern DAC when approaching Nyquist; I don't have a strong belief one way or the other over which would win, given that 14-bits of dynamic range is about where the integrity of your analog path starts to matter a whole lot.
[edit]
Sound Blaster claims "up to" 130dB of dynamic range on theirs. Putting aside how much work the "up to" id doing, this would make the last 9 or 10 bits completely useless, even for undithered signals, but I was thinking of something cheaper, by one of those fly-by-night companies with names made of random letters.
R&S equipment is pretty common in the cellular industry. There is more of it out there than you would think. While this particular signal generator is maybe not the most common piece of equipment you will see lots of R&S spectrum analyzers and the like.
It's technically not a signal generator but an Arbitrary Waveform Generator. It provides a base-band modulated signal to the signal generator which the sig gen upconverts to the RF carrier. If you purchased a Vector Signal Generator this would be built in but you can still buy them stand alone and they are pretty common. NI, R&S, Tek, and Keysight all have product lines of them.
Another way to think about it, this would be comparable to a signal generator in the same way that an oscilloscope is comparable to a spectrum analyzer.
What you call a signal generator is what most would call an RF signal generator. Both an AWG and an RF signal generator belong to the generic signal generator family. :-)
I'm trying really hard not to want to turn this into a dac for my stereo system. They build really nice boxes..
I have a PSL-3 industrial controller (well, two, but one is broken) on the shelf behind me that I'm using for the noble purpose of playing Theme Hospital in Windows 98.
It's not impossible to hack the hardware to do this: the output of the SDRAMs go to some 74-series chips which drive both the DAC and the connector for the 16-bit output.
You could remove the SDRAM path altogether, reverse the signal path from the connector with some PCB surgery and drive the DAC from the connector with your own digital hardware. That's one way...
There's also a hardware path from the ISA bus straight to the DAC, bypassing the SDRAM, for diagnostic purposes probably. So there might by options there too, but though it would require a ton of reverse engineering or even reprogramming the FPGAs (which are so old that the software isn't available anymore on the Altera website.)
Whether or not all of that is worth doing for a 14-bit DAC output is a matter of personal opinion.
A 14-bit dac at 105msamples/s with delta sigma modulation means you can get an effective bit depth of ~50 bits @20khz which is a ridiculous amount of dynamic range. Far more than what is implied by "just" 14-bits.
I'm amused that it's just a PC motherboard, and that it uses ISA to talk to the generator hardware itself. Makes perfect sense; nothing needs high data rates there, and it's really easy to interface with. I just wouldn't have necessarily predicted it before seeing the case open.
Many modern measurement hardware is built this way - motherboard with custom PCIe boards. It dramatically simplifies GUI design using standard tools and keeps the complexity separated on the external boards. The PCIe boards also can be designed for one/two measurement channels making the equipment rather modular and customizable.
A lot of hardware is built like this. Separating the GUI from the real time or safety-critical aspects of the system is a common enough need that there are many System-on-Module (SoM) boards that can run Linux or Windows and also include an ARM processor for real time behavior on the same die.
Your UI can be built in Python/Qt/Tk while the safety critical stuff is programmed in C running on an RTOS.
I meant safety critical, although a similar design decision is often made for time critical behavior.
In some cases, e.g., for IEC-62304 compliance, software is designated as various safety classes based on likelihood of causing Harm. If you can extract the more safety critical software from the rest of the system and prove that it is sufficiently segregated (e.g., by putting it on a separate processor), you can substantially reduce your Verification and documentation burden.
I haven't done it myself... yet... but these days it's not that horrible to DIY PCIe with an FPGA. There's even a fully open source PCIe core, and there are example designs for it: https://github.com/enjoy-digital/litepcie.
Wonderful article and I don't even know that this device existed perhaps due to Agilent/Keysight bias laboratory/shop.
R & S AMIQ Modulation Generator general specs is available here [1].
Fun facts, the venerable Silicon Valley was started by Hewlett-Packard (HP, then Agilent, now Keysight) that mainly supplying function/signal/waveform generator for the then booming radar and electronics industry.
HP Garage is now a designated landmark and marked with a plaque calling it the "Birthplace of 'Silicon Valley' " [2].
Their first customer was actually Disney, who needed a wide-range audio oscillator to test sound hardware that would be used in theaters showing 'Fantasia.'
The oscillator's model number was 200A, because even back then they understood the SV maxim "Nobody wants to buy the first version of anything."
You'll find whole labs full of it once in a while on Bidspotter, set a saved-search for some names like Rohde and Tektronix and Keysight and you'll get emails once in a while.
Extremely good deals are rare and the chance that there’s something broken is very real. I often consider that a feature: fixing test equipment is a bit of a hobby, but be prepared to end up with a doorstop that cost a few hundred dollars.
Here are some examples that I bought though them:
- HP 8650E spectrum analyzer: works but intermittent shutdowns
- TDS 794D: works, but CRT is broken. A well know failure mode. Need to install a $75 LCD replacement.
- this R&S AMIQ: totally broken. Needed many days of work to revive.
Great disassembly job. There's yet another branch of the rabbit hole that Tom almost fell down:
The 'translinear' school of design was founded (or at least popularized) by Barrie Gilbert at Analog Devices (https://en.wikipedia.org/wiki/Translinear_circuit , https://en.wikipedia.org/wiki/Barrie_Gilbert ). Gilbert was one of the undisputed GOATs in the analog field and a full-fledged Renaissance man in his own right. It's worth reading more about him if/when you have the time and inclination.Gilbert was a legend of analog design.
Here's a lecture by him on the translinear principle: https://youtu.be/LQNJVtcFrCc
And here's some history of his work at Tektronix (where he discovered the translinear principle): https://vintagetek.org/barrie-gilbert/
That’s for that info! I’ll give that links a good read.
It would be interesting to benchmark this against consumer USB DACs that are marketed for sound. You can get 32-bit 192kHz bandwidth (384kHz sampling rate) DACs now for under $200. Obviously this is much lower bandwidth than the R&S, but I'd be interested to see the dynamic range of a modern DAC when approaching Nyquist; I don't have a strong belief one way or the other over which would win, given that 14-bits of dynamic range is about where the integrity of your analog path starts to matter a whole lot.
[edit]
Sound Blaster claims "up to" 130dB of dynamic range on theirs. Putting aside how much work the "up to" id doing, this would make the last 9 or 10 bits completely useless, even for undithered signals, but I was thinking of something cheaper, by one of those fly-by-night companies with names made of random letters.
Some of this obscure test equipment is very interesting and amazingly designed.
There are probably only a few electrical engineers in the entire world that actually use this stuff on a day to day basis.
R&S equipment is pretty common in the cellular industry. There is more of it out there than you would think. While this particular signal generator is maybe not the most common piece of equipment you will see lots of R&S spectrum analyzers and the like.
It's technically not a signal generator but an Arbitrary Waveform Generator. It provides a base-band modulated signal to the signal generator which the sig gen upconverts to the RF carrier. If you purchased a Vector Signal Generator this would be built in but you can still buy them stand alone and they are pretty common. NI, R&S, Tek, and Keysight all have product lines of them.
Another way to think about it, this would be comparable to a signal generator in the same way that an oscilloscope is comparable to a spectrum analyzer.
What you call a signal generator is what most would call an RF signal generator. Both an AWG and an RF signal generator belong to the generic signal generator family. :-)
I'm trying really hard not to want to turn this into a dac for my stereo system. They build really nice boxes..
I have a PSL-3 industrial controller (well, two, but one is broken) on the shelf behind me that I'm using for the noble purpose of playing Theme Hospital in Windows 98.
It's not impossible to hack the hardware to do this: the output of the SDRAMs go to some 74-series chips which drive both the DAC and the connector for the 16-bit output.
You could remove the SDRAM path altogether, reverse the signal path from the connector with some PCB surgery and drive the DAC from the connector with your own digital hardware. That's one way...
There's also a hardware path from the ISA bus straight to the DAC, bypassing the SDRAM, for diagnostic purposes probably. So there might by options there too, but though it would require a ton of reverse engineering or even reprogramming the FPGAs (which are so old that the software isn't available anymore on the Altera website.)
Whether or not all of that is worth doing for a 14-bit DAC output is a matter of personal opinion.
A 14-bit dac at 105msamples/s with delta sigma modulation means you can get an effective bit depth of ~50 bits @20khz which is a ridiculous amount of dynamic range. Far more than what is implied by "just" 14-bits.
I'm amused that it's just a PC motherboard, and that it uses ISA to talk to the generator hardware itself. Makes perfect sense; nothing needs high data rates there, and it's really easy to interface with. I just wouldn't have necessarily predicted it before seeing the case open.
Many modern measurement hardware is built this way - motherboard with custom PCIe boards. It dramatically simplifies GUI design using standard tools and keeps the complexity separated on the external boards. The PCIe boards also can be designed for one/two measurement channels making the equipment rather modular and customizable.
A lot of hardware is built like this. Separating the GUI from the real time or safety-critical aspects of the system is a common enough need that there are many System-on-Module (SoM) boards that can run Linux or Windows and also include an ARM processor for real time behavior on the same die.
Your UI can be built in Python/Qt/Tk while the safety critical stuff is programmed in C running on an RTOS.
I think you meant 'time critical'.
I meant safety critical, although a similar design decision is often made for time critical behavior.
In some cases, e.g., for IEC-62304 compliance, software is designated as various safety classes based on likelihood of causing Harm. If you can extract the more safety critical software from the rest of the system and prove that it is sufficiently segregated (e.g., by putting it on a separate processor), you can substantially reduce your Verification and documentation burden.
Yeah, but PCIe is a bit harder to DIY an interface for! ;)
I haven't done it myself... yet... but these days it's not that horrible to DIY PCIe with an FPGA. There's even a fully open source PCIe core, and there are example designs for it: https://github.com/enjoy-digital/litepcie.
Wonderful article and I don't even know that this device existed perhaps due to Agilent/Keysight bias laboratory/shop.
R & S AMIQ Modulation Generator general specs is available here [1].
Fun facts, the venerable Silicon Valley was started by Hewlett-Packard (HP, then Agilent, now Keysight) that mainly supplying function/signal/waveform generator for the then booming radar and electronics industry.
HP Garage is now a designated landmark and marked with a plaque calling it the "Birthplace of 'Silicon Valley' " [2].
[1] Rohde & Schwarz – AMIQ04 I/Q Modulation Generator:
https://testworld.com/product/rohde-schwarz-amiq04-iq-modula...
[2] Hewlett-Packard:
https://en.wikipedia.org/wiki/Hewlett-Packard
Their first customer was actually Disney, who needed a wide-range audio oscillator to test sound hardware that would be used in theaters showing 'Fantasia.'
The oscillator's model number was 200A, because even back then they understood the SV maxim "Nobody wants to buy the first version of anything."
Does anyone know what company auctions this stuff? The author doesn't say.
You'll find whole labs full of it once in a while on Bidspotter, set a saved-search for some names like Rohde and Tektronix and Keysight and you'll get emails once in a while.
I had the same question. Seems the UC schools use Publicsurplus.com, but last time I looked, I didn't see anything interesting.
I bought it through edispositions.com.
Extremely good deals are rare and the chance that there’s something broken is very real. I often consider that a feature: fixing test equipment is a bit of a hobby, but be prepared to end up with a doorstop that cost a few hundred dollars.
Here are some examples that I bought though them:
- HP 8650E spectrum analyzer: works but intermittent shutdowns
- TDS 794D: works, but CRT is broken. A well know failure mode. Need to install a $75 LCD replacement.
- this R&S AMIQ: totally broken. Needed many days of work to revive.
It’s high risk buying these things.
ELI5: what is this used for?
It’s used to create communication test signals.
Say you designed a Bluetooth transceiver chip and need to test the performance under stress conditions: noise, sample clock instability etc.
The WinIQSim software simulates such a signal and uploads it to the AMIQ, which is essentially a high quality digital analog converter.
The analog output will go to an RF signal generator and then to your chip under test.