SNR differences between REW and QA40x using the QA403

[stef1777]

Hi John,

I've been comparing REW 5.40 beta 135 with QA40x 1.224, using the same QA403 to measure an amplifier at approximately 1 W into 8 Ω.

I performed two separate comparisons:

1. SNR measurements (generator ON)

With a nominal 1 kHz test signal at 192 kHz, QA40x reported an SNR of 99.08 dB, while REW reported 94.4 dB, a difference of approximately 4.7 dB. The measured amplifier gain was nearly identical: 27.45 dB and 27.46 dB, respectively.

2. Noise measurements (generator OFF)

With the generator disabled but still physically connected, QA40x measured −89.14 dBV and REW measured −90.17 dBV, a difference of only 1.03 dB.

These are separate measurements: the generator-OFF readings represent noise levels in dBV, whereas the SNR figures express the ratio between signal and noise.

Importantly, the original SNR comparison was made before we discovered the ASIO sample rate startup issue. REW was actually generating 250 Hz instead of 1 kHz, so those SNR results need to be repeated at the correct frequency.

Once the sample rate is verified, I'd like to understand whether REW and QA40x use different SNR calculation methods or measurement bandwidths that could explain any remaining difference.

Thanks!

 

[stef1777]

Noise.

 

[John Mulcahy]

The important settings for REW are the distortion high pass and low pass and whether "Use AES17-2015 standard notch" is selected. If it is, there's a standard one octave notch centred on the fundamental. If it isn't, REW follows the main lobe of the fundamental to determine what part of the signal belongs to it. Select "Highlight fundamental" to see what's included.

 

[stef1777]

I'll look into this tomorrow. I also need to redo the QA403 calibrations in REW, as they were affected by the sample rate issue, which I wasn't aware of at the time.

 

[stef1777]

Hi John,

Following your suggestions, I repeated the measurements on October 24. All screenshots are included in the attached ZIP.

Settings:

• QA403, REW 5.40 beta 135 / ASIO401 2.0, QA40x 1.224
• Q17 Mini amplifier, 8 ohm load
• 192 kHz, 256k FFT, 32 averages, 20 Hz–20 kHz, no weighting
• QA403 input range: +18 dBV
• Generator ON: 1 kHz, 2.89 Vrms measured at the amplifier output in both applications
• Generator OFF: same setup, amplifier still connected

Observations (generator ON, Rectangular window):

REW: SNR 88.3 dB, THD 0.0012%, THD+N 0.0040%, N -79.1 dBV

QA40x: SNR 99.01 dB, THD 0.00116%, THD+N 0.00169%, N-D -90.14 dBV

AES17 notch ON/OFF changed REW's SNR by only 0.1 dB. Hann versus Rectangular also made little difference.

Generator OFF:

• REW: N -86.4 dBV
• QA40x: N-D -92.11 dBV

The actual 1 kHz frequency was verified with the oscilloscope. An intermediate QA40x test with an incorrect input range was discarded.

The SNR difference remains 10.71 dB, despite nearly identical THD results.

Could this be explained by differences in SNR/noise calculation methods, or am I missing a setting?

Thanks again for your help!

 

[John Mulcahy]

REW is using the ratio of the fundamental power to the noise plus non-harmonic distortion power as SNR. Both are unweighted.

 

[stef1777]

Could the approximately 11 dB difference be explained by how REW and QA40x exclude spectral components from their noise calculations?

 

[John Mulcahy]

Perhaps, I'm looking at it just now to make sure there are no numerical issues for low noise and high THD.

 

[stef1777]

I hope you can get to the bottom of this, whether the issue is with REW or QA40x. I'm starting to feel a bit out of my depth here!

 

[John Mulcahy]

I've simulated your results as closely as I can and I get the same answers as REW is showing. I have also checked the REW calculations using signals with noise floors approx 40 dB lower and the results remain correct, though I can improve their precision a little by changing the way the N+NHD summation is done which I'll do in the next build. It has no bearing on your results though. The QA40x THD+N looks too low and SNR looks too high to me given where the harmonics are relative to the noise floor, unless there is some other factor in play, such as A weighting, or I'm just wrong