Hi Chris,
Like I said earlier, I do not mind helping out on keeping dlbSVO going, but you know more of this design than I do and so does RNM. RNM has one but he says he can not get his hands on it.
So I take it, that David's proto's and RNM's are the only ones in existence unless others step up to say otherwise.
I have asked a few questions about it and I have not heard any answers, so RNM is the only one that can answer these questions, unless we get David's computer and project related items to evaluate. Questions on the table are:
1) The control interface details. Yes I know it is SPI, but what drives it other than saying the obvious, a computer, signals and wires
Was there a USB to parallel port bridge involved? Something like,
FT221X
2) Performance details as Bob has asked today.
Maybe these answers are buried in this thread, for which someone has to dig and find out. Some detective work is required to start with.
Rick
Like I said earlier, I do not mind helping out on keeping dlbSVO going, but you know more of this design than I do and so does RNM. RNM has one but he says he can not get his hands on it.
So I take it, that David's proto's and RNM's are the only ones in existence unless others step up to say otherwise.
I have asked a few questions about it and I have not heard any answers, so RNM is the only one that can answer these questions, unless we get David's computer and project related items to evaluate. Questions on the table are:
1) The control interface details. Yes I know it is SPI, but what drives it other than saying the obvious, a computer, signals and wires
Was there a USB to parallel port bridge involved? Something like,FT221X
2) Performance details as Bob has asked today.
Maybe these answers are buried in this thread, for which someone has to dig and find out. Some detective work is required to start with.
Rick
Last edited:
Hi Chris,
Sounds like to me that you will be forced to get into the "nitty gritty" details in order to sort out/find where all the information resides on his computer.
While you are at it, ask for the project bits and pieces (parts, proto's,docs), if they can identify or locate them.
Good Luck
Sounds like to me that you will be forced to get into the "nitty gritty" details in order to sort out/find where all the information resides on his computer.
While you are at it, ask for the project bits and pieces (parts, proto's,docs), if they can identify or locate them.
Good Luck
As I said before, I supported his work in spirit, direction and financially. Though I had Victors gen.... I didnt want to buy several of them and asked david if he would like to work on one that I could choose any freq I wanted via software interface (USB).
So, I have one completed unit. Here is David's harmonics..... a bit noisy environment but it is very low level ....
However, this also shows a low level calibration tone added at exactly -100dbv of higher freq to the gen signal. The analyzer locks onto the higher level fundamental tone, only. So, zero on the screen is -100. I have fiddled with internal control/trim pots and get typically -145 for harmonics.
With David's suggestion it could be lower. -160 could be possible.
GUI Software control is done also.
THx-RNMarsh
So, I have one completed unit. Here is David's harmonics..... a bit noisy environment but it is very low level ....
However, this also shows a low level calibration tone added at exactly -100dbv of higher freq to the gen signal. The analyzer locks onto the higher level fundamental tone, only. So, zero on the screen is -100. I have fiddled with internal control/trim pots and get typically -145 for harmonics.
With David's suggestion it could be lower. -160 could be possible.
GUI Software control is done also.
THx-RNMarsh
Last edited:
Getting back on track
I would like to return to the topic at hand with a couple of thoughts and questions on oscillators.
I have been toying with the idea of implementing a very low distortion oscillator, inspired by Victor's oscillator and the discussions on this thread.
Its been almost 40 years since I designed a new low-distortion oscillator, namely the SVO in my THD analyzer. It could do about -120 dB at 1 kHz and about -110 dB at 20 kHz, using NE5534 op amps. Pretty good at the time, and it has served my needs well. That was inspired in part by Bruce Hofer's AES paper on the SVO used in the Tek SG505 while he was still at Tek.
I like to think in terms of 3 types of oscillators. A dedicated fixed-frequency one like Victor's, a switched frequency category like that in the HP339A, and a continuously-variable class of oscillators.
It is my belief that it is easier to obtain very low distortion in a fixed-frequency oscillator because everything can be optimized for that frequency, including precision of tuning components. Precision of tuning components leads to ability to use less invasive agc, and hence lower distortion from the agc level detector and control device (e.g., JFET).
Switched-frequency designs are a bit more difficult to achieve ultra-low distortion, but they are more versatile. They have a higher cost due to the much larger number of passive tuning components and the cost of the switching. There is also more opportunity for parasitic and other effects to degrade performance. Granularity of frequency selection is also an important consideration.
Oscillators like that in the HP 339A have quite fine granularity. At the other extreme, a switched frequency oscillator that only performed at 50 Hz, 1 kHz and 20 kHz would serve a great many needs with adequate versatility. 90% of the distortion tests I do are at one of those 3 frequencies. A middle ground for a fixed-frequency oscillator might be one that covers 4 decades in a 1-2-5 sequence, perhaps starting at 5 Hz and ending at 50 kHz. Each decade might actually have 4 frequencies in a 5-10-20-50 sequence, for example, providing overlap at decade edges for user convenience. This is still within range of reasonable complexity and just a couple of small-signal relays right on the board could do the required switching of the R's and C's. I think the whole range can be covered by just 4 DPDT relays for R and 4 relays for C's. Frequency selection could be done by either a pair of simple one-gang 5-position rotary switches or by a processor that controls the relays.
Truly continuously-tuned oscillators over each decade are usually the most difficult to implement with ultra-low distortion performance, partly because of tracking issues in the variable tuning element, be it a dual pot or a dual air capacitor. Once again, less precise tracking of the tuning elements generally leads to the need for more aggressive agc circuits and hence more distortion.
What are the thoughts here on needed degree of granularity in an ultra-low distortion switched frequency oscillator?
Cheers,
Bob
I would like to return to the topic at hand with a couple of thoughts and questions on oscillators.
I have been toying with the idea of implementing a very low distortion oscillator, inspired by Victor's oscillator and the discussions on this thread.
Its been almost 40 years since I designed a new low-distortion oscillator, namely the SVO in my THD analyzer. It could do about -120 dB at 1 kHz and about -110 dB at 20 kHz, using NE5534 op amps. Pretty good at the time, and it has served my needs well. That was inspired in part by Bruce Hofer's AES paper on the SVO used in the Tek SG505 while he was still at Tek.
I like to think in terms of 3 types of oscillators. A dedicated fixed-frequency one like Victor's, a switched frequency category like that in the HP339A, and a continuously-variable class of oscillators.
It is my belief that it is easier to obtain very low distortion in a fixed-frequency oscillator because everything can be optimized for that frequency, including precision of tuning components. Precision of tuning components leads to ability to use less invasive agc, and hence lower distortion from the agc level detector and control device (e.g., JFET).
Switched-frequency designs are a bit more difficult to achieve ultra-low distortion, but they are more versatile. They have a higher cost due to the much larger number of passive tuning components and the cost of the switching. There is also more opportunity for parasitic and other effects to degrade performance. Granularity of frequency selection is also an important consideration.
Oscillators like that in the HP 339A have quite fine granularity. At the other extreme, a switched frequency oscillator that only performed at 50 Hz, 1 kHz and 20 kHz would serve a great many needs with adequate versatility. 90% of the distortion tests I do are at one of those 3 frequencies. A middle ground for a fixed-frequency oscillator might be one that covers 4 decades in a 1-2-5 sequence, perhaps starting at 5 Hz and ending at 50 kHz. Each decade might actually have 4 frequencies in a 5-10-20-50 sequence, for example, providing overlap at decade edges for user convenience. This is still within range of reasonable complexity and just a couple of small-signal relays right on the board could do the required switching of the R's and C's. I think the whole range can be covered by just 4 DPDT relays for R and 4 relays for C's. Frequency selection could be done by either a pair of simple one-gang 5-position rotary switches or by a processor that controls the relays.
Truly continuously-tuned oscillators over each decade are usually the most difficult to implement with ultra-low distortion performance, partly because of tracking issues in the variable tuning element, be it a dual pot or a dual air capacitor. Once again, less precise tracking of the tuning elements generally leads to the need for more aggressive agc circuits and hence more distortion.
What are the thoughts here on needed degree of granularity in an ultra-low distortion switched frequency oscillator?
Cheers,
Bob
For myself, there is a trade quality vs cost. When I see a circuit Davida oscillator, according to other participants " five star" quality in terms of thd & granularity (16-bits dac per decade frequency or so), I try to weight it vs >50$ (CAD) per each of those 16-bits dacs (there are like 3 of them). And I start to think, that probably I don't need 65.000 steps frequency per decade at that price. Alternative is tlc7541 12-bits for 1/4 price, or even DG413 for 1/25. 16 steps still not so bad, and 4096 steps more than I would ever need. I 'd try tlc7528 first, dual for 1/8 compare to Davida's, not sure about THD, DS says -85dBc vs -103dBc
It appears to me that the main interest in recent weeks (or even months) is in a fourth type, an oscillator that's (voltage or otherwise automatically) tunable over a very small range around a common "standard" frequency such as 999.5Hz to 1000.5Hz while achieving as low distortion as feasible with this feature. This allows syncing with a digital clock signal that drives a D/A converter to generate data for an FFT, as recently discussed.
David's oscillator has several interesting and novel features. The AGC which can be the most challenging was done with a multiplying DAC and an ADC. The ADC can be used to sample the output precisely to maintain amplitude accuracy and a multiplying DAC can be as linear as any solution for level stability. I proposed to David some time ago using some logic between the DAC and the ADC to implement fast start and fast settling. It would be a small extension to also use the output of the ADC and some logic to drive another multiplying DAC for frequency trim/stabilization.
Another direction would be an active state variable filter on the output of a DAC either driven directly from a PC or running a sine from a file on a rom chip. Implemented with the multiplying DAC's and feedback to reduce distortion. It would be very like an injection stabilized oscillator.
Unfortunately there are no cheap solutions to multiplying DAC but the LTC2731 claims -110 dB THD for around $10. The package is a challenge but not insurmountable.
Another direction would be an active state variable filter on the output of a DAC either driven directly from a PC or running a sine from a file on a rom chip. Implemented with the multiplying DAC's and feedback to reduce distortion. It would be very like an injection stabilized oscillator.
Unfortunately there are no cheap solutions to multiplying DAC but the LTC2731 claims -110 dB THD for around $10. The package is a challenge but not insurmountable.
I would support this. Having a very low distortion oscillator is one thing, surely a very important thing, the start of it all. But also important is what you do with it, how you can measure your DUT's output. With standard analog tests you can reliably maybe measure down to -140dBc, -150dBc on a good day. So if you can design a -160dBc or better oscillator, you need to look at the other side of the chain.
I believe that if you want to measure down to -160dBc or lower you need to go to synchronous and averaged FFT. For that, you must be able to lock the oscillator to a ref frequency.
In other words, I would prefer a -160dBc lockable oscillator to a -180dBc free running one.
My € 0.02 worth.
Jan
Bob,
When I use a fully sweepable signal source it is being used to test for resonances. When I need increadably low residual distortion it is to do distortion testing.
I am unaware of resonant distortion.
Of course my preference is looking at intermodulation distortion. With decent selection of the test frequencies the intermodulation result is quite different than the test signal harmonics.
The biggie for really low level tests for me has always been AC power line noise.
Of course to do a sweepable signal generator my approach would be to start with generating a sine wave and perhaps 9-15 of the harmonics and use sine/cosine lock-in amplifiers to determine the phase and levels required to produce a low distortion sine. Implemented in analog this could be done with reasonably repeated and scaled semi-complex circuitry.
In digital it would actually be simpler. Two D/A converters scaled perhaps 0 dB and -120 dB. Lots of synchronous sample and holds and a multiplexed A/D or so.
Reminds me looking at another month of more idle time than normal, I need to order a new microprocessor education kit.
When I use a fully sweepable signal source it is being used to test for resonances. When I need increadably low residual distortion it is to do distortion testing.
I am unaware of resonant distortion.
Of course my preference is looking at intermodulation distortion. With decent selection of the test frequencies the intermodulation result is quite different than the test signal harmonics.
The biggie for really low level tests for me has always been AC power line noise.
Of course to do a sweepable signal generator my approach would be to start with generating a sine wave and perhaps 9-15 of the harmonics and use sine/cosine lock-in amplifiers to determine the phase and levels required to produce a low distortion sine. Implemented in analog this could be done with reasonably repeated and scaled semi-complex circuitry.
In digital it would actually be simpler. Two D/A converters scaled perhaps 0 dB and -120 dB. Lots of synchronous sample and holds and a multiplexed A/D or so.
Reminds me looking at another month of more idle time than normal, I need to order a new microprocessor education kit.
> That's sooo 19th century ;-)
If you happen to have 150 Million € and what to buy one of the latest EUV scanners from a certain Dutch company,
they have stabilisation time of more than 12 hours.
And that is State of the Art. There ain't anything better.
Sometimes you cannot change the law of physics.
Patrick
If you happen to have 150 Million € and what to buy one of the latest EUV scanners from a certain Dutch company,
they have stabilisation time of more than 12 hours.
And that is State of the Art. There ain't anything better.
Sometimes you cannot change the law of physics.
Patrick
Honestly I cannot imagine fine-tuning an analog ultra-low distortion oscillator down to fractions of Hz for long FFTs. Short FFTs will require constant measuring the momentary frequency, adaptive resampling to fit the bins for averaging, and many runs for averaging . The resampling must be of high quality to not introduce artefacts of its own (trivial to simulate this effect with SoX with its configurable resampling and high-resolution spectrogram output). That means specific software support and a substantial computing power. Both readily available to anyone basically for free in the 21st century.
DA/AD converters can be quite easily compensated for harmonic distortion in the digital domain, again takes just software and a simple hardware. The DAC signal can be cleaned for non-harmonic artefacts with a higher-order passive low-distortion band-pass filter. This cleaning is not required for distortion measurements if the fundamental is chosen to not overlap with the loop artefacts (why measuring with 1kHz when that is the most common artefact in USB-based gear?). If the cleaning is required, the band-pass filter can be re-tuned for the given frequency with a feedback controlled by the software, again quite easily and inexpensively attainable.
What does this require? A quality DA/AD gear with standard (or at least known) communication protocol so that new software functions can be added. Such hardware can be quite pricey, but if the gear can be extended with new software features at any time, its value can last for decades. RTX6001 uses standard USB-audio. QA401 does have an open source (i.e. long-term supportable) ASIO driver. Most other gear is proprietary and users are at mercy of the vendors who will never implement all the specialities users may need now and in the future. Actually the most likely result is that they will abandon their legacy product with a new version of windows.
Coding new features in octave/Julia/Python/Csound/PureData is quite simple, there are many people who can do it right away, all the tools are legally available for free, with huge community, constantly receiving updates and improvements. The code can be shared (github), a library could be built up (most of the functions are already available anyway, it takes just putting them into a workflow). IMO fusing minimum required hardware and maximum software, produced and shared by the community of users, is the 21st century. Example being the software-defined radio community, achieving incredible results with decently-priced hardware and vast collection of open software.
That does not diminish the importance of the hardware parameters. No software can separate DUT noise from measurement noise (sum of white noises is a white noise again), increase samplerate to gain more information, reliably compensate distortion down to tiny levels if the hardware itself distorts a lot or is unstable, etc.
IMO RTX6001 or QA401 or anything not existing yet with faster sampling would do wonders when accompanied with custom DSP in playback and capture streams. Anyone can try it, play with it, use it. They can start with their existing soundcards and upgrade later on if needed. All it takes is just willingness to learn new stuff.
DA/AD converters can be quite easily compensated for harmonic distortion in the digital domain, again takes just software and a simple hardware. The DAC signal can be cleaned for non-harmonic artefacts with a higher-order passive low-distortion band-pass filter. This cleaning is not required for distortion measurements if the fundamental is chosen to not overlap with the loop artefacts (why measuring with 1kHz when that is the most common artefact in USB-based gear?). If the cleaning is required, the band-pass filter can be re-tuned for the given frequency with a feedback controlled by the software, again quite easily and inexpensively attainable.
What does this require? A quality DA/AD gear with standard (or at least known) communication protocol so that new software functions can be added. Such hardware can be quite pricey, but if the gear can be extended with new software features at any time, its value can last for decades. RTX6001 uses standard USB-audio. QA401 does have an open source (i.e. long-term supportable) ASIO driver. Most other gear is proprietary and users are at mercy of the vendors who will never implement all the specialities users may need now and in the future. Actually the most likely result is that they will abandon their legacy product with a new version of windows.
Coding new features in octave/Julia/Python/Csound/PureData is quite simple, there are many people who can do it right away, all the tools are legally available for free, with huge community, constantly receiving updates and improvements. The code can be shared (github), a library could be built up (most of the functions are already available anyway, it takes just putting them into a workflow). IMO fusing minimum required hardware and maximum software, produced and shared by the community of users, is the 21st century. Example being the software-defined radio community, achieving incredible results with decently-priced hardware and vast collection of open software.
That does not diminish the importance of the hardware parameters. No software can separate DUT noise from measurement noise (sum of white noises is a white noise again), increase samplerate to gain more information, reliably compensate distortion down to tiny levels if the hardware itself distorts a lot or is unstable, etc.
IMO RTX6001 or QA401 or anything not existing yet with faster sampling would do wonders when accompanied with custom DSP in playback and capture streams. Anyone can try it, play with it, use it. They can start with their existing soundcards and upgrade later on if needed. All it takes is just willingness to learn new stuff.
Last edited:
“ Reminds me looking at another month of more idle time than normal, I need to order a new microprocessor education kit.”
Look at the mbed ARM platform. Common C++ platform that covers most ARM derivatives from the big guys. Loads of fantastic on-chip peripherals and you get fast 32 bit computing power.
Home | Mbed
If you need any pointers, just shout.
Look at the mbed ARM platform. Common C++ platform that covers most ARM derivatives from the big guys. Loads of fantastic on-chip peripherals and you get fast 32 bit computing power.
Home | Mbed
If you need any pointers, just shout.
> That's sooo 19th century ;-)
If you happen to have 150 Million € and what to buy one of the latest EUV scanners from a certain Dutch company,
they have stabilisation time of more than 12 hours.
And that is State of the Art. There ain't anything better.
Sometimes you cannot change the law of physics.
Patrick
All the more reason to go for synchronous lock. Frequency drift (wrt the measurement equipment clock) is then by definition zero as is the stabilization time.
BTW, € 150M? I would think in these times they would give me a discount ;-)
Jan
Precise results require very precise calculations (64bit double float) often not available to low-power embedded platforms. In such case an alternative could be a full-blown linux platform, e.g. RPi. Most DSP functions are already coded (python, julia, octave), it takes just chaining them + the plumbing around. Plus very easy development and debugging with monitor/keyboard/networking etc. Large community behind every feature/function, easy and free access to code authors down to almost every part.
Unless hard-realtime hardware timing is required, of course.
Unless hard-realtime hardware timing is required, of course.
I find this a very important and interesting discussion. We are at the threshold of making a big step forward in measuring distortion to unprecedented low levels. But it is clear that analog-only solutions won't cut it; mixed analog-digital or better yet all-digital solutions are the way to go.
That however requires other technical knowledge and expertise than what I grew up with.
I'm all for learning new tricks, but reality forces me to acknowledge that, being in my 8th decade, learning gets harder and harder, and slower and slower. I hate getting old!
Jan
That however requires other technical knowledge and expertise than what I grew up with.
I'm all for learning new tricks, but reality forces me to acknowledge that, being in my 8th decade, learning gets harder and harder, and slower and slower. I hate getting old!
Jan
Last edited: