This thread describes a simple interface to an M-Audio Audiophile-192 (AP-192) balanced soundcard for measuring the distortion of balanced and unbalanced amplifiers. Equivalent results should be obtainable using other high-quality soundcards. Although the results shown are for the Pass F5 unbalanced amplifier, the same circuits are applicable to balanced amplifiers.
Background:
About a year ago I built multiple channels of the Nelson Pass F5 amplifier, but had no objective way (other than my ear) to measure their performance. I built a Cordell Distortion Magnifier (DM) and installed an M-Audio Audiophile-192 soundcard on my Linux machine. After some modifications to the DM, I had very good results tweaking the F5 performance. But, I discovered that the AP-192 by itself could measure distortions to below .001%, without needing the Distortion Magnifier.
Relevent Specifications of the AP-192:
See M-AUDIO - Audiophile 192 - High-Definition 4-In/4-Out Audio Card with Digital I/O and MIDI
. 24-bit, 192KHz max
. Balanced inputs and outputs
. Max input voltage: 4 V rms (5.66 V peak)
. Input impedence: 20K balanced, 10k unbalanced.
. Max output voltage: 1.95 RMS unbal (2.8V peak)
In order to protect the AP-192 soundcard input from overload due to the full output voltage of the amplifier, a 4:1 resistive attenuator is placed between the amplifier output and the soundcard input.
Requirements for testing the F5:
. Unbalanced inputs and outputs
. Maximum output signal of 20V peak for 25W into 8 ohms.
. Amplifier Gain: 6x
(Note: 6 x 2.8V = 16.8V attainable from AP-192 output)
Note that the F5 channel being tested was built using Cviller V1.1 boards and matched Toshiba 2SK1530 and 2SJ201 MOSFETs. This F5 channel also contains the P3 potentiometer described in http://www.diyaudio.com/forums/pass-labs/121228-f5-power-amplifier-1086.html#post2667669, which was adjusted for minimum 2nd harmonic distortion.
Shown below are schematics for both a balanced and unbalanced amplifier, and a spectrum measurement of the F5 output at 1.05KHz 1 watt output into 8 ohms.
Final warning: Since there is no protection against overloading your soundcard input, be careful to provide an appropriate attenuation ratio the resistor dividers.
Background:
About a year ago I built multiple channels of the Nelson Pass F5 amplifier, but had no objective way (other than my ear) to measure their performance. I built a Cordell Distortion Magnifier (DM) and installed an M-Audio Audiophile-192 soundcard on my Linux machine. After some modifications to the DM, I had very good results tweaking the F5 performance. But, I discovered that the AP-192 by itself could measure distortions to below .001%, without needing the Distortion Magnifier.
Relevent Specifications of the AP-192:
See M-AUDIO - Audiophile 192 - High-Definition 4-In/4-Out Audio Card with Digital I/O and MIDI
. 24-bit, 192KHz max
. Balanced inputs and outputs
. Max input voltage: 4 V rms (5.66 V peak)
. Input impedence: 20K balanced, 10k unbalanced.
. Max output voltage: 1.95 RMS unbal (2.8V peak)
In order to protect the AP-192 soundcard input from overload due to the full output voltage of the amplifier, a 4:1 resistive attenuator is placed between the amplifier output and the soundcard input.
Requirements for testing the F5:
. Unbalanced inputs and outputs
. Maximum output signal of 20V peak for 25W into 8 ohms.
. Amplifier Gain: 6x
(Note: 6 x 2.8V = 16.8V attainable from AP-192 output)
Note that the F5 channel being tested was built using Cviller V1.1 boards and matched Toshiba 2SK1530 and 2SJ201 MOSFETs. This F5 channel also contains the P3 potentiometer described in http://www.diyaudio.com/forums/pass-labs/121228-f5-power-amplifier-1086.html#post2667669, which was adjusted for minimum 2nd harmonic distortion.
Shown below are schematics for both a balanced and unbalanced amplifier, and a spectrum measurement of the F5 output at 1.05KHz 1 watt output into 8 ohms.
Final warning: Since there is no protection against overloading your soundcard input, be careful to provide an appropriate attenuation ratio the resistor dividers.
Attachments
Sorry, but I forgot to mention that all of my measurements are made using the Sillanumsoft Visual Analyser Visual Analyser 2011 XE software, which runs natively on Windows and on Linux using Wine.
Interpreting THD measurements
Some of you who know better are suspicious (or should be) of the .0007% THD measurement shown in the spectrum plot in post #1. I have made distortion measurements of the same F5 board using the Cordell Distortion Magnifier (DM) and measure a THD of around .0011%. Both sets of measurements are repeatable. I do not believe that these results are due to random measurement error.
Here is an partial explanation for difference in the THD measurements:
The M-Audio AP-192 DAC(D-to-A converter) and ADC (A-to-D converter) introduce harmonic distortion to analog chain. The M-Audio spec sheet provides ADC THD=.0006% and DAC THD=.0023%, at 1 KHz and 2 V RMS signal levels.
When a signal is passes thru a chain of distorting elements, each component in the chain contributes, in a complex fashion, to final output signal. Thus if we pass a pure sine wave of amplitude X[f] at frequency f, thru a distorting element with harmonic coefficients H, we get signal whose spectrum is:
Y[i*f] = H * X[f]
Consider a chain of 2 distorting components with coefficients H1 and H2. If H/H[1] are very small for i>1, then second order distortions such as produced when the the harmonic i*f passes thru H2[j] for i>1 an
H3 = H1 + H2
What is critical to understand is that the harmonic coefficients H are COMPLEX NUMBERS. They represent both a scaling and phase-shift of the signal. Depending on the phase-shifts, distortions in the component chain can reinforce or cancel one another.
BRACKETING THE AMPLIFIER THD:
To determine the error bounds on the amplifier THD measurement, you must measure both the loopback THD
of the soundcard, and the total THD of the DAC, AMP, and DAC analog chain.
THD_total - THD_lb <= THD_amp <= THD_total + THD_lb
Shown below are 4 measured spectra for the F5 board. They provide the soundcard loopback measurement and
combined amplifier measurements using the soundcard interface in post #1, and using the Cordell Distortion Magnifier (DM).
MEASUREMENTS USING THE CORDELL DISTORTION MAGNIFIER:
The DM measurements were made using x10 magnification, where the harmonic components (other than the fundamental) are multiplied by x10 (+20db).
Thus loopback THD of .0010 becomes .0001.
THD_loop = .0001
THD_total = .0011
.0010% <= THD_amp <= .0012
Soundcard caused distortions are attenuated by a factor of 10 using the DM in x10 mode, thus THD_total is
mostly influenced bt the distortions of the amplifier.
MEASUREMENTS USING THE RESISTOR DIVIDER INTERFACE OF POST #1:
THD_loop = .0006
THD_total = .0007
.0001% <= THD_amp <= .0013
Thus using the interface of post #1, it appears that the soundcard distortions are out-of-phase with the F5 amplifier distortions, thus causing significant harmonic cancellation.
I hope this helps to explain how to interpret the THD measurements.
Some of you who know better are suspicious (or should be) of the .0007% THD measurement shown in the spectrum plot in post #1. I have made distortion measurements of the same F5 board using the Cordell Distortion Magnifier (DM) and measure a THD of around .0011%. Both sets of measurements are repeatable. I do not believe that these results are due to random measurement error.
Here is an partial explanation for difference in the THD measurements:
The M-Audio AP-192 DAC(D-to-A converter) and ADC (A-to-D converter) introduce harmonic distortion to analog chain. The M-Audio spec sheet provides ADC THD=.0006% and DAC THD=.0023%, at 1 KHz and 2 V RMS signal levels.
When a signal is passes thru a chain of distorting elements, each component in the chain contributes, in a complex fashion, to final output signal. Thus if we pass a pure sine wave of amplitude X[f] at frequency f, thru a distorting element with harmonic coefficients H, we get signal whose spectrum is:
Y[i*f] = H * X[f]
Consider a chain of 2 distorting components with coefficients H1 and H2. If H/H[1] are very small for i>1, then second order distortions such as produced when the the harmonic i*f passes thru H2[j] for i>1 an
H3 = H1 + H2
What is critical to understand is that the harmonic coefficients H are COMPLEX NUMBERS. They represent both a scaling and phase-shift of the signal. Depending on the phase-shifts, distortions in the component chain can reinforce or cancel one another.
BRACKETING THE AMPLIFIER THD:
To determine the error bounds on the amplifier THD measurement, you must measure both the loopback THD
of the soundcard, and the total THD of the DAC, AMP, and DAC analog chain.
THD_total - THD_lb <= THD_amp <= THD_total + THD_lb
Shown below are 4 measured spectra for the F5 board. They provide the soundcard loopback measurement and
combined amplifier measurements using the soundcard interface in post #1, and using the Cordell Distortion Magnifier (DM).
MEASUREMENTS USING THE CORDELL DISTORTION MAGNIFIER:
The DM measurements were made using x10 magnification, where the harmonic components (other than the fundamental) are multiplied by x10 (+20db).
Thus loopback THD of .0010 becomes .0001.
THD_loop = .0001
THD_total = .0011
.0010% <= THD_amp <= .0012
Soundcard caused distortions are attenuated by a factor of 10 using the DM in x10 mode, thus THD_total is
mostly influenced bt the distortions of the amplifier.
MEASUREMENTS USING THE RESISTOR DIVIDER INTERFACE OF POST #1:
THD_loop = .0006
THD_total = .0007
.0001% <= THD_amp <= .0013
Thus using the interface of post #1, it appears that the soundcard distortions are out-of-phase with the F5 amplifier distortions, thus causing significant harmonic cancellation.
I hope this helps to explain how to interpret the THD measurements.
Attachments
My configuration is M-Audio AP-192 and Sillanumsoft Visual Analyser running under Wine on Linux. For some reason, the anti-aliasing shelf filters of the DAC and ADC are being set to 20 KHz for all sample rates, thus making useful harmonic analysis impossible above about about 6 KHz fundamental.
Because of the way that the Cordell Distortion Magnifier does its nulling of the fundamental test frequency, distortion components from the DAC are also subtracted. This means that the loopback measurements thru the DM are measuring mainly the distortions due to the ADC. This suggests that the difference between the loopback measured without and with the DM might be a reasonable estimate of the DAC THD.
Using the data given above, that gives THD_DAC in the range .0005 to .0007.
OTOH, the M-Audio spec sheet says that the DAC THD+NOISE is .0023% at 1 KHz -1db below full output.
Bottom line: Without a super-duper low distortion signal generator, I cannot be sure.
lhquam,
Thanks for the info and the thread. Just a few days ago I started this thread http://www.diyaudio.com/forums/software-tools/198886-thd-im-software-suggestions.html#post2756269
I was looking at RealRTA, then I saw ARTA and now you are using Visual Analyzer and there is SMAART.
Have you used anything other than Visual Analyzer?
I read your post about the Cordell DM, but it sounds like it is truly not necessary. I'm still going through all of your data and I appreciate your input.
Thanks,
Dave
Thanks for the info and the thread. Just a few days ago I started this thread http://www.diyaudio.com/forums/software-tools/198886-thd-im-software-suggestions.html#post2756269
I was looking at RealRTA, then I saw ARTA and now you are using Visual Analyzer and there is SMAART.
Have you used anything other than Visual Analyzer?
I read your post about the Cordell DM, but it sounds like it is truly not necessary. I'm still going through all of your data and I appreciate your input.
Thanks,
Dave
From wat i understand, the oscillator in this thread makes for a very good investment if youare doing a lot of testing. It is based on AN-67.
http://www.diyaudio.com/forums/equi...erter-project-audio-measurements-tool-11.html
Here is thrread with more info
http://www.diyaudio.com/forums/grou...alow-thd-10khz-sine-oscillator-group-buy.html
http://www.diyaudio.com/forums/equi...erter-project-audio-measurements-tool-11.html
Here is thrread with more info
http://www.diyaudio.com/forums/grou...alow-thd-10khz-sine-oscillator-group-buy.html
Last edited:
Question if I may: is anyone of you using their soundcard with W7 64bit? I have an older EMU 0404 external soundcard box, but can't get it to run under W7 64bit. Several forums document this; EMU does have a beta driver for W7 64bit but it seems to work for some, not for others, without clear info why.
Maybe I should get another like the M-Audio 192kHz, but would like to be sure that it would run under W7 64bit.
BTW I have a license for ARTA/STEPS.
TIA,
jan didden
Maybe I should get another like the M-Audio 192kHz, but would like to be sure that it would run under W7 64bit.
BTW I have a license for ARTA/STEPS.
TIA,
jan didden
The salient question that may be poised by members at this stage is how much distortion was truly superimposed by the testing apparatus (IO soundcard & DAC) over the amplifier being tested, and how do we compensate for it because there are several values shown? Another member tried this method and found much different results. The measured values vary from that on the CDM accordingly to each device's own distortion character and order proportions and final interference of each. How would each compare to that shown on reliable scientific calibrated outboard spectrum analyzers that properly compensate for the test probes signal reduction? From what I'm seeing, the data displayed using this sound card method is prone to error and difficult to interpret without proper calibration. So, how should users actually interpret the results for their own perusal?
Last edited:
Kouiky:
By doing loopback tests of the M-Audio AP-192 I found that noise and distortion was much reduced using balanced outputs and inputs rather than unbalanced. My impression is that until amplifier being tested (DUT) has noise and distortion levels approaching those of the soundcard, the soundcard can be largely ignored.
Since the soundcard interface presented here is only intended for measurements of the low impedance amplifier output (usually 8 ohms or less), and is of a totally resistive nature, there should be no calibration issue other than overall signal level. It you accurately know the resistor divider and the soundcard levels you should be good, otherwise you need to calibrate by some means. For THD measurement, the absolute signal level accuracy is less important than the ratios of harmonic levels to fundamental.
By doing loopback tests of the M-Audio AP-192 I found that noise and distortion was much reduced using balanced outputs and inputs rather than unbalanced. My impression is that until amplifier being tested (DUT) has noise and distortion levels approaching those of the soundcard, the soundcard can be largely ignored.
Since the soundcard interface presented here is only intended for measurements of the low impedance amplifier output (usually 8 ohms or less), and is of a totally resistive nature, there should be no calibration issue other than overall signal level. It you accurately know the resistor divider and the soundcard levels you should be good, otherwise you need to calibrate by some means. For THD measurement, the absolute signal level accuracy is less important than the ratios of harmonic levels to fundamental.
This is a good read from a BSEE who worked in the audio industry for years. It's good for anyone trying to test equipment via USB, soundcard and freeware and explains the limitations of reliability due to unseen interactions within a computer that the software and card can't account for and common errors. Interestingly, it appears many audio manufacturers still print erroneous performance results because they are using these methods. He tested a company's product and after the manufacturer saw his results, they used his values! The high end manufacturer didn't even have test equipment and their headphone amp almost destroyed the headphones.
NwAvGuy: RightMark Audio Analyzer
NwAvGuy: RightMark Audio Analyzer
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.