Update in the interface- changing the output requirement and putting it into the box really helped simplify it. I have figured out that I only need +/- 10V supplies to make it work and that the power requirements of the chips are small so I'll change the supplies to a charge pump setup and I won't need to isolate the grounds. Saves about $20 from the BOM as well. Don't need to go to surface mount I hope.
On the Shibasoku the noise dominates the reading at about .0009% THD+N. The distortion is around -115 or .00015% for 2-5. The low operating level (1V) is what is limiting things but it should be more than adequate for real work.
On the Shibasoku the noise dominates the reading at about .0009% THD+N. The distortion is around -115 or .00015% for 2-5. The low operating level (1V) is what is limiting things but it should be more than adequate for real work.
The "D" version -- AD725D -- has a 10db lower distortion level that can be switched in.... the .0003% full scale (fs) range. Or, -110dB fs.
I can see -130db OK on that range. The monitor output shows that there is ability to show lower than -130dB. The 725D, at the monitor output port... shows to -150dB.
Didnt you say the unit you have was good to -140dB?
Confused.
THx-RNMarsh
I can see -130db OK on that range. The monitor output shows that there is ability to show lower than -130dB. The 725D, at the monitor output port... shows to -150dB.
Didnt you say the unit you have was good to -140dB?
Confused.
THx-RNMarsh
Last edited:
I have just finished the PCB layout and expect to send it for manufacturing tomorrow (I will just check it once more before sending). If all goes well I should have the PCB prototypes in a couple of weeks.
Before designing the interface I measured the clock signals of the USBStreamer, including some of the internal signals around the PLL that generates the MCLK, and I think I have figured out how it works.
Then I need to get some of the components, mount a couple of boards and test it.
I hope that I will also be able to source some 22.5792MHz crystal oscillators in a 3.2 x 2.5 mm package. I have not been able to find them in small quantities at Farnell, Mouser etc. 24.576MHz is no problem.
The "D" version -- AD725D -- has a 10db lower distortion level that can be switched in.... the .0003% full scale (fs) range. Or, -110dB fs.
I can see -130db OK on that range. The monitor output shows that there is ability to show lower than -130dB. The 725D, at the monitor output port... shows to -150dB.
Didnt you say the unit you have was good to -140dB?
Confused.
THx-RNMarsh
The older 725's have a similar feature but 10 dB less sensitivity. These numbers show the many ways of measuring and reporting distortion and noise.
Earlier isolators where discussed, now I found this one. Can anyone tell me if this will work (or has a good probability of working ).
http://hjelmslund.dk/GET/PDF-USBISO-AA.asp
).http://hjelmslund.dk/GET/PDF-USBISO-AA.asp
It has limited functionality, since it is only a full speed device. High speed is not supported. And it only solves the isolation part of it, not the generation of a clean clock signal. And the DC/DC converter used internall might create a noise problem.
I guess it is OT, but the aim of my design is to:
1. Be able to provide a clean master clock and let the PC lock to that.
2. Isolate from the computer with minimal coupling between the two sides.
3. Do the above on a unit, which supports I2S outputs AND I2S inputs.
Some existing designs may do 1 and 2, but I have not seen one that does all of them.
And what if the clock on the interface is not clean enough? Then it is just a matter of providing a "perfect clock" from an external clock source.
Regarding number 3 my inputs will probably end up having a left justified format instead of I2S, but that should not normally be a big problem.
I guess it is OT, but the aim of my design is to:
1. Be able to provide a clean master clock and let the PC lock to that.
2. Isolate from the computer with minimal coupling between the two sides.
3. Do the above on a unit, which supports I2S outputs AND I2S inputs.
Some existing designs may do 1 and 2, but I have not seen one that does all of them.
And what if the clock on the interface is not clean enough? Then it is just a matter of providing a "perfect clock" from an external clock source.
Regarding number 3 my inputs will probably end up having a left justified format instead of I2S, but that should not normally be a big problem.
testing - V2.0, preliminary
So.
After way too much time involved with the process of installing a fresh copy of Ex Pee on an HP laptop. You don't really want to ask... it is sufficient to say that for whatever reasons the company that puts out the Ex-Pee product (etc) seems to revel in making almost everything as obscure and difficult as possible *except* for when it works properly as expected. Which is some times and not others. This was one of those not others.
Finally, a shot while ago all the magic incantations properly invoked and the internal electrons tamed and forced to run together in formation, I installed the QA400 on what is now my "measurement only" laptop.
Unfortunately, I find myself once again at distance from the VP-7725, so I can't do the check with the oscillator of that unit vs. the internal oscillator.
However, I did do some quick checks with the QA in loopback.
This time the results were somewhat at difference from the first time!
Quieter in general.
This may be significant, as I dropped the oscillator output from (uncalibrated) -10db to -13db I saw in both channels a drop of ~10dB in the "spurious" harmonics that were noted earlier. So, this appears to be non-linear.
The total dynamic range to the noise floor (with averaging up about 10) appears to the eye to be ~125dB, this with the oscillator input at about that point where the spurs drop into the noise.
The other thing I noticed is that the LF noise floor varied as the cables (this is in loopback) and the unit was moved around and touched. Undoubtedly there is somewhat less ambient electrical noise in this location compared to the former. But this theory will have to be confirmed to be certain.
No screen shots this time, as the program I use to reduce file size has not yet migrated to the measurement laptop.
So, to summarize, the differences between channels appear to be much less today, nearly nil except to the extent that the oscillators appear to be producing spurs as the level gets above -10dB. I will have to look more closely at this relationship, and double check to see if the left/right oscillator (cable) swap reveals anything about the source of the spurs - is it overload/non-linearity at the input of the measurement side, or at the output of the oscillator??
Tune in next time for another thrill packed episode of Distortion Sleuths!
_-_-
So.
After way too much time involved with the process of installing a fresh copy of Ex Pee on an HP laptop. You don't really want to ask... it is sufficient to say that for whatever reasons the company that puts out the Ex-Pee product (etc) seems to revel in making almost everything as obscure and difficult as possible *except* for when it works properly as expected. Which is some times and not others. This was one of those not others.
Finally, a shot while ago all the magic incantations properly invoked and the internal electrons tamed and forced to run together in formation, I installed the QA400 on what is now my "measurement only" laptop.
Unfortunately, I find myself once again at distance from the VP-7725, so I can't do the check with the oscillator of that unit vs. the internal oscillator.
However, I did do some quick checks with the QA in loopback.
This time the results were somewhat at difference from the first time!
Quieter in general.
This may be significant, as I dropped the oscillator output from (uncalibrated) -10db to -13db I saw in both channels a drop of ~10dB in the "spurious" harmonics that were noted earlier. So, this appears to be non-linear.
The total dynamic range to the noise floor (with averaging up about 10) appears to the eye to be ~125dB, this with the oscillator input at about that point where the spurs drop into the noise.
The other thing I noticed is that the LF noise floor varied as the cables (this is in loopback) and the unit was moved around and touched. Undoubtedly there is somewhat less ambient electrical noise in this location compared to the former. But this theory will have to be confirmed to be certain.
No screen shots this time, as the program I use to reduce file size has not yet migrated to the measurement laptop.
So, to summarize, the differences between channels appear to be much less today, nearly nil except to the extent that the oscillators appear to be producing spurs as the level gets above -10dB. I will have to look more closely at this relationship, and double check to see if the left/right oscillator (cable) swap reveals anything about the source of the spurs - is it overload/non-linearity at the input of the measurement side, or at the output of the oscillator??
Tune in next time for another thrill packed episode of Distortion Sleuths!
_-_-
Hi Bear ---- your results are consistent with mine... non-linear behavior leads to errors in determining the DUT's 2h and 3h and others. And, I also see changes with time and temperature on top of the non-linearity's. -10 to -13dB input level beginning to go non-linear leaves a more limited dynamic range for useful DUT testing.
For very low THD tests, it looks like this unit, as well as sound cards and the like, must use some form or fundamental suppression to sqeeze out more range and accuracy. [But that range also needs to be tested for accuracy etc ]. Those spurious spikes are a real nuisance when trying to know what is produced by the DUT.
Thx-RNMarsh
For very low THD tests, it looks like this unit, as well as sound cards and the like, must use some form or fundamental suppression to sqeeze out more range and accuracy. [But that range also needs to be tested for accuracy etc ]. Those spurious spikes are a real nuisance when trying to know what is produced by the DUT.
Thx-RNMarsh
Last edited:
RNM,
Quite so.
But determining which end, or both(?) the non-linearity is coming from may be important. I'll know more when I get to try the VP-7725's oscillator on it.
I know Demian ran some tests earlier with an external oscillator, and it looks like the non-linearity is in the input stages going to the A/D or in the A/D from what he got... which is not good. Maybe a better opamp inside the box for the incoming signal??
Also this AM I was thinking that I wonder how they change the oscillator level, and if that is having some sort of effect on the distortion.
The other idea is that a ~10dB amp on the output of the oscillator (assuming it is itself ultra low distortion) might be sufficient to eek out the last 10dB of dynamic range from the QA?
Otoh, if this is on the input of the AD or the amp preceding it a somewhat different approach is needed.
What would be nice is an inverse "cal file" method to subtract the spurious stuff before a test. One would do a loopback at the requisite level, and then after the measurement, the software might subtract the residuals, (via the "cal file" just made) giving a spur free result. A bit of tail chasing, but maybe useful.
Overall, I'm ok with the QA at the moment. But one definitely needs to pay close attention to the cabling, the environmental RFI and the like. I'm thinking that perhaps the lower noise is also due to the computer running next to nothing more than the OS, and may serendipity in the form of a lower noise USB port. Last time running on batts made no difference compared to using the PS.
_-_-
Quite so.
But determining which end, or both(?) the non-linearity is coming from may be important. I'll know more when I get to try the VP-7725's oscillator on it.
I know Demian ran some tests earlier with an external oscillator, and it looks like the non-linearity is in the input stages going to the A/D or in the A/D from what he got... which is not good. Maybe a better opamp inside the box for the incoming signal??
Also this AM I was thinking that I wonder how they change the oscillator level, and if that is having some sort of effect on the distortion.
The other idea is that a ~10dB amp on the output of the oscillator (assuming it is itself ultra low distortion) might be sufficient to eek out the last 10dB of dynamic range from the QA?
Otoh, if this is on the input of the AD or the amp preceding it a somewhat different approach is needed.
What would be nice is an inverse "cal file" method to subtract the spurious stuff before a test. One would do a loopback at the requisite level, and then after the measurement, the software might subtract the residuals, (via the "cal file" just made) giving a spur free result. A bit of tail chasing, but maybe useful.
Overall, I'm ok with the QA at the moment. But one definitely needs to pay close attention to the cabling, the environmental RFI and the like. I'm thinking that perhaps the lower noise is also due to the computer running next to nothing more than the OS, and may serendipity in the form of a lower noise USB port. Last time running on batts made no difference compared to using the PS.
_-_-
Before going too far off the deep end check the datasheet for the codec in the middle of the QA400 http://www.cirrus.com/en/pubs/proDatasheet/CS4272_F1.pdf It really defines what is possible. I have been able to get -110 THD loopback with it and the external box. It seems a low Z drive helps. The top 5 dB will give a degraded performance but below that you still have enough dynamic range and low enough distortion to be able to analyze and verify pretty much any real world audio device I know of.
Grounding and even radiated noise will degrade the performance. I need to either run on battery or keep the power brick for my laptop well away from the QA400 when making these types of measurements.
If you need lower my suggestion would be to build a clone of the Shibasoku notch filter and follow that with an amplifier. The 40 dB band reject will give an effective -140 dB+ distortion floor with not too critical tuning.
Grounding and even radiated noise will degrade the performance. I need to either run on battery or keep the power brick for my laptop well away from the QA400 when making these types of measurements.
If you need lower my suggestion would be to build a clone of the Shibasoku notch filter and follow that with an amplifier. The 40 dB band reject will give an effective -140 dB+ distortion floor with not too critical tuning.
All of the designs I play with are below -100dB. So, clean grounds, power supplies and test equipment to below -120 is needed as a minimum. Even this stock $35K A-P FFT isn't good enough for my experimental circuits. It will need some help (notch) but it is clean and no PS or grounding issues etc. And can take a wide range of input level without distortion and noise issues/spurious in the way.
[And, it can do a bewildering number of different kinds of tests.]
When I need to see below the 24 bit level, the THD test mode is better than the 24 bit FFT (if the noise is low enough).
But thank the Gods for the ShibaSoku as I can read THD/THD+N and individual harmonics like I would expect to. Plug-N-Play.
Thx-RNMarsh
[And, it can do a bewildering number of different kinds of tests.]
When I need to see below the 24 bit level, the THD test mode is better than the 24 bit FFT (if the noise is low enough).
But thank the Gods for the ShibaSoku as I can read THD/THD+N and individual harmonics like I would expect to. Plug-N-Play.
Thx-RNMarsh
Last edited:
Thanks for that heads up Demian.
What do you think of Cordell's "Distortion Magnifier" as an alternate approach here?
I'll try to digest the pdf a bit later on...
Overall I am pretty happy with this box as is it will certainly handle most of what needs to be measured, it will take a very high performance DUT to exceed its limitations. In which case I at least can still switch to the VP-7725 and eak out a bit more dynamic range. But I am still interested in the Shibasoku idea.
_-_-
What do you think of Cordell's "Distortion Magnifier" as an alternate approach here?
I'll try to digest the pdf a bit later on...
Overall I am pretty happy with this box as is it will certainly handle most of what needs to be measured, it will take a very high performance DUT to exceed its limitations. In which case I at least can still switch to the VP-7725 and eak out a bit more dynamic range. But I am still interested in the Shibasoku idea.
_-_-
I wish I had a distortion magnifier to play with. Its a great idea and used with the QA400 might be a great tool to get deeper. Without one I can't really comment further.
On my never ending to-do list is the Shibasoku filter for the box.
I made a lot of progress on figuring how to get the interface board into the box but have had little time to check and finish the details. The nice part is that 1/4" TRS jacks can fit the holes the BNC connectors are in so it becomes a pretty clean mod. I found a charge pump that will generate +/- 15V from 5V and is in a dip package so its all real close. Just need to finish the design.
Maybe next week.
On my never ending to-do list is the Shibasoku filter for the box.
I made a lot of progress on figuring how to get the interface board into the box but have had little time to check and finish the details. The nice part is that 1/4" TRS jacks can fit the holes the BNC connectors are in so it becomes a pretty clean mod. I found a charge pump that will generate +/- 15V from 5V and is in a dip package so its all real close. Just need to finish the design.
Maybe next week.
I put together the "Distortion Magnifier" and put it in a box etc. Note - you need to have the DUT have a gain of 10 or more for it to work well. --> cant use it on buffers and lower gain circuits (filters, headphone amps etc).
Otherwise, it is a very good idea but the notch is more universally useful.
[ PS.. I'm not using the Magnifier.]
Thx-RNMarsh
Otherwise, it is a very good idea but the notch is more universally useful.
[ PS.. I'm not using the Magnifier.]
Thx-RNMarsh
Last edited:
Here are a few results from today.
VP-7725 driving the input of the QA.
Set the 7725 as close as possible to the "bin" that the QA selects for "1kHz" outside that bin, the fundamental's shape gets wider on the display. That was 1.002kHz, iirc. QA called for 1.00198 or something like that.
Forgot to run the output of the QA's oscillator into the VP-7725.
Oy!
The VP7725 *may* be overly optimistic about what it reads. I did buy this one as-is surplus. Perhaps, perhaps not. Oddly the external loop via BNC coax measures just slightly better than the internal (push a button) loopthrough.
I was also thinking that the LF carp that the QA seems to pick up through the case (maybe it *needs or wants* to be grounded externally? Are the inputs and output BNCs floating - forgot to look at that?) is factored into the number that the QA puts up as THD?
They do not agree about the ratios of harmonics, but the QA does have that residual spike stuff going on, but it seems to be ok down to right at 0.001%. Going above about -10dB on the input of the QA with the VP7725 oscillator driving it still increases the spikies.
My LEFT channel is still somewhat cleaner than the right.
Neither has been "calibrated" yet. And they do not agree on levels, so that needs doing.
At this point I'd be surprised (but that happens too often anyhow) if the A/D is doing the spikey bit, so I am thinking, maybe it is the input opamp(s)? If so, perhaps surgical intervention is called for? That extra 10dB would be sweet.
Oh, the residual with nothing plugged in drops the noise floor down to ~-130dB cleanly.
I looked in vain for a BNC dummy load or just a BNC cap to cover off the unused BNCs, couldn't find one anywhere! Gah! I *know* they are lurking and mocking me.
There is definite feed through/crosstalk between channels of the QA. Although I am not sure if it is internal or the BNCs talking to each other yet. (no I did not make up a shorting or dummy load BNC, time was at a high premium...)
The images shown are with the QA oscillators off. Things are not as pretty with them on - I ran them at higher unrelated frequencies just to see if it showed, it does.
That's the report for the Distortion Sleuths for today.
VP-7725 driving the input of the QA.
Set the 7725 as close as possible to the "bin" that the QA selects for "1kHz" outside that bin, the fundamental's shape gets wider on the display. That was 1.002kHz, iirc. QA called for 1.00198 or something like that.
Forgot to run the output of the QA's oscillator into the VP-7725.
Oy!
The VP7725 *may* be overly optimistic about what it reads. I did buy this one as-is surplus. Perhaps, perhaps not. Oddly the external loop via BNC coax measures just slightly better than the internal (push a button) loopthrough.
I was also thinking that the LF carp that the QA seems to pick up through the case (maybe it *needs or wants* to be grounded externally? Are the inputs and output BNCs floating - forgot to look at that?) is factored into the number that the QA puts up as THD?
They do not agree about the ratios of harmonics, but the QA does have that residual spike stuff going on, but it seems to be ok down to right at 0.001%. Going above about -10dB on the input of the QA with the VP7725 oscillator driving it still increases the spikies.
My LEFT channel is still somewhat cleaner than the right.
Neither has been "calibrated" yet. And they do not agree on levels, so that needs doing.
At this point I'd be surprised (but that happens too often anyhow) if the A/D is doing the spikey bit, so I am thinking, maybe it is the input opamp(s)? If so, perhaps surgical intervention is called for? That extra 10dB would be sweet.
Oh, the residual with nothing plugged in drops the noise floor down to ~-130dB cleanly.
I looked in vain for a BNC dummy load or just a BNC cap to cover off the unused BNCs, couldn't find one anywhere! Gah! I *know* they are lurking and mocking me.
There is definite feed through/crosstalk between channels of the QA. Although I am not sure if it is internal or the BNCs talking to each other yet. (no I did not make up a shorting or dummy load BNC, time was at a high premium...)
The images shown are with the QA oscillators off. Things are not as pretty with them on - I ran them at higher unrelated frequencies just to see if it showed, it does.
That's the report for the Distortion Sleuths for today.
Attachments
Last edited: