Agreed, the Asus Essence STX would be a better choice, but don't forget, I started this project ten years ago. By that time, only a Lynx ($$$) was better.
As a matter of fact it's a lousy sound card, apparently designed by a 'digital' guy. Wrong schematic, wrong op-amps, even wrong resistors. Aargh!
I've extensively modified the card, which resulted in a ~10dB better performance.
Regarding the DAC output, for critical measurements I put a passive bandpass filter behind it. Furthermore, the ADC signal is massively averaged and finally, to lower the measurement floor, a 'reference sine' is subtracted from it. This 'reference sine' is obtained by using the same setup, though without DUT (of course taking care of the same amplitude of the DAC and ADC). As a result, the measurement floor has been brought down to -140...-145dB.
As for spurious noise, not just averaging (millions of samples) takes care of it, also outliers (spikes) are detected and are automatically skipped.
Cheers,
E.
Source for this info? Or care to post some results? Are you sure you mean a dynamic range of 118dB?
I can't imagine any reason why a practical implementation of PCM1794A could degrade the dynamic range with more than 10dB under the datasheet spec.
"I can't imagine any reason why a practical implementation of PCM1794A could degrade the dynamic range with more than 10dB under the datasheet spec."
You have got to kidding with a statement like that.
You have got to kidding with a statement like that.
Unless, of course, the designers intentionally degraded the noise floor by a factor of 3. There are PCM179xA reference designs on the TI site, with layouts, etc... Even the "audiophile" PCM chinese clones available on EBay are doing better than 118dB dynamic range.
I think the scope for error at -118 to -127 dB is significant. You are talking about <500nV ref 1 V out to <1 uV at 2 V out. It takes very little to introduce errors of tt magnitude.
By measuring the D/A noise floor, using a 1KHz notch filter, a spectrum analyzer and averaging, then calculating the RMS noise. 20log(V1KHz/VRMSnoise) is the dynamic range. E.g. for a 5V 1KHz output and 5uV noise floor, the dynamic range is 120dB.
The notch filter attenuation adds to the spectrum analyzer dynamic range, so measuring 5uV is very easy in a good lab. We do such measurements in our university lab on a daily basis.
A test by Techreport , including swapping op amps :
Asus' Xonar DG and Xense sound cards - The Tech Report - Page 1
The DAC is a PCM1796 still a 132dB dynamic , according to TI,
the 1794 being obsolete.
Numbers are quite good , though , and should be excellent
using averaging.
Quote "By measuring the D/A noise floor, using a 1KHz notch filter, a spectrum analyzer and averaging, then calculating the RMS noise. 20log(V1KHz/VRMSnoise) is the dynamic range. E.g. for a 5V 1KHz output and 5uV noise floor, the dynamic range is 120dB."
I am quite aware of that of how to calculate the dynamic range.
Read my post again. That is not what I asked.
Clearly your comprehension needs to improve considerably if you are to qualify as a half useful engineer or physicist.
I am quite aware of that of how to calculate the dynamic range.
Read my post again. That is not what I asked.
Clearly your comprehension needs to improve considerably if you are to qualify as a half useful engineer or physicist.
Last edited:
Set aside the pesonal touch you brought to the discussion, I have to admit I don't understand what you are asking here.
I hope you are not confusing the output signal to noise ratio (118dB in the table on the first page) with the dynamic range. Otherwise, the dynamic range measurements using rightmark are showing a much better performance, even @44KHz.
So where is that 118dB dynamic range you quoted coming from?
My bad , i did confuse with the Realtek numbers..
Anyway , THD is quite high in respect of the dynamic range ,
even if it s measured only at 1Khz , so amps THD measurements
are not specialy easier than with Edmond s card.
Dynamic range is not all...
Anyway , THD is quite high in respect of the dynamic range ,
even if it s measured only at 1Khz , so amps THD measurements
are not specialy easier than with Edmond s card.
Dynamic range is not all...
Once that enough dynamic range is provided, a DSA instrument can measure the spectral components both in magnitude and phase. Then normalizing by substracting the input and signal processing residuals is easy and the nulled output result may have residuals at under -140dB.
But I agree, I am not aware of any sound card that would measure under 1ppm out of the box. That's why I was asking about Edmond's method, and I'm still not convinced it is capable to cover such a huge difference in the dynamic range (of only -106dB).
BTW, to measure the spectral component phase you obviously need a reference channel, not sure how to implement such with a sound card, perhaps by using the second (left or right) channel? But are the sound card channels phase coherent? For a DSA instrument, they are guaranteed to be! I have no idea, but such questions convinced me some time ago that using a sound card for measuring 1ppm distortions is not realistic.
I'm not saying that a DSA should be on every DIY Audio member. But before making extraordinary, simulation based only, claims about 0.00001% distortions, you'd better make sure you are able to support such claims by a consistent, calibrated and cross-checked for reproductibility measurement methodology.
sound cards can be used with indirect techniques - generate multitone sines in separate channels, add together ("no distortion" inverting summing amp or even just some Rs) then filter or otherwise "remove" high amplitude test signals from DUT output, amplify and ADC
no -120 dB distortion ADC or DAC needed - my ESI Juli@ setup gave ~160 dB noise floor limited distortion resolution in a measurement with 100 s averaging
while my setup didn't require cancellation in the output I believe that is a good approach - you don't need 100+ dB null of the test signal in the output - just knock it down enough that the ADC's own distortion products with the test signal residuals are low enough - after all with today's free tools we are going to look with a fft not just staring at some rms voltmeter dial any more
not so easy to sweep, may need multiple different sines/tests for different distortion orders
but hard to match the resolution - can easily beat AP at the cost of flexibility - have to test (each) spot frequency, harmonic order
no -120 dB distortion ADC or DAC needed - my ESI Juli@ setup gave ~160 dB noise floor limited distortion resolution in a measurement with 100 s averaging
while my setup didn't require cancellation in the output I believe that is a good approach - you don't need 100+ dB null of the test signal in the output - just knock it down enough that the ADC's own distortion products with the test signal residuals are low enough - after all with today's free tools we are going to look with a fft not just staring at some rms voltmeter dial any more
not so easy to sweep, may need multiple different sines/tests for different distortion orders
but hard to match the resolution - can easily beat AP at the cost of flexibility - have to test (each) spot frequency, harmonic order
Yes, top of the range AP comes in at 50k IIRC. I signed off the budget for one for the Tokyo lab 2 or 3 years ago. The old one was written off but someone got to it before me. Bastard.
If i understand , the principle is to filter the amp s output signal
from fundamental frequency with a filter , wich will let the HD contents
at the same amplitude , hence extending the measurable range by as much as
the attenuation ratio of the filter, assuming this latter has very low THD itself.
from fundamental frequency with a filter , wich will let the HD contents
at the same amplitude , hence extending the measurable range by as much as
the attenuation ratio of the filter, assuming this latter has very low THD itself.
by looking for the IMD sum or difference frequencies you don't need the individual sines to have low harmonic distortion - ordinary soundcard DAC output is fine as long as you only make one of each sine in the multone test in its own DAC channel
since the IMD amplitude is proportional to the product of the amplitude of the exciting waves even 40 dB attenuation of the test sines in the output to your final ADC reduces the ADC generated distortion by an "extra" ~80 dB over its full scale distortion spec
since the IMD amplitude is proportional to the product of the amplitude of the exciting waves even 40 dB attenuation of the test sines in the output to your final ADC reduces the ADC generated distortion by an "extra" ~80 dB over its full scale distortion spec
Another possibility to increase measurement dynamic range for simple signal (one sine),without filters, is to use full duplex soundcard, and to generate the same signal in both channels, but one channel inverted (delayed exactly one period, it is possible with some measurement programs ). Output of one channel to input of DUT, and mix DUT output signal with inverted signal from soundcards second channel , with correct amplitude, to obtain maximum supression of fundamental frequency. To some extent it can supress soundcard generator distortion, too (supposing both output channels are equivalent)
Last edited:
Measuring IMD with sound cards
Hi JCX,
You are quit right about the DAC channels for a multi tone test, that is, the two tones should be generated in separate DACs (of course). But sometimes that is not enough. I did a 19/20kHz ITU-R IMD test with my waveterminal 192X. Without attenuation of the test sines, the 1kHz component measures 102.4dB lower. Next, with 20dB attenuation of the test sines, the 1k component was not much lower, it only dropped to -107.1dB. So something went wrong with the generation of the sines and not with the acquisition of them. Whether it stems from an interaction between the DACs (sitting on the same chip) or the rest of the circuit, I don't know exactly. BUT..... when I generate the two sines in separate sound cards, results are much better: without attenuator 1k measures -109.1dB and with the attenuator in place, I get -136.6dB, being an improvement of ~30dB. (btw, the averaging time was 43 sec.)
Conclusion: Your ESI Juli card do a much better job.
I also wonder how today's high-end sound cards (from asus) perform in this respect, i.e. is one card sufficient for low level IMD measurements, or do you need two cards?
Cheers,
E.
Hi JCX,
You are quit right about the DAC channels for a multi tone test, that is, the two tones should be generated in separate DACs (of course). But sometimes that is not enough. I did a 19/20kHz ITU-R IMD test with my waveterminal 192X. Without attenuation of the test sines, the 1kHz component measures 102.4dB lower. Next, with 20dB attenuation of the test sines, the 1k component was not much lower, it only dropped to -107.1dB. So something went wrong with the generation of the sines and not with the acquisition of them. Whether it stems from an interaction between the DACs (sitting on the same chip) or the rest of the circuit, I don't know exactly. BUT..... when I generate the two sines in separate sound cards, results are much better: without attenuator 1k measures -109.1dB and with the attenuator in place, I get -136.6dB, being an improvement of ~30dB. (btw, the averaging time was 43 sec.)
Conclusion: Your ESI Juli card do a much better job.
I also wonder how today's high-end sound cards (from asus) perform in this respect, i.e. is one card sufficient for low level IMD measurements, or do you need two cards?
Cheers,
E.
- Status
- Not open for further replies.