D/A preamp FFT measurements for review plz

[zigzagflux]

Finally was able to get some decent measurements of the distortion of my equipment. Up to this point I had a Tek scope with FFT, but the noise floor was around -55 dB and was only able to detect gross problems. Now I am using a Focusrite Saffire FW box with Arta and Audiotester. Haven't decided which package to purchase yet; still under probation.

At least all the measurements I have obtained with both softwares are identical, so I believe my procedure to be reasonably correct. Now, I would like to know if my results are good/bad/average.

My preamp schematic is attached. Not shown is the K&K Audio Mark II DAC, with SPDIF input and differential current output PCM1794. The I/V is shown feeding the differential gainstage.

Three FFT's are attached. No calibrations were performed. First one is a loopback using soundcard analog out, just to see how the output looks. Second is output of the entire preamp at max volume. Third is output of the D/A taken at the grids, full volume.

Questions are hopefully easy to answer (sorry for the scaling differences):

1. For the loopback, what would be the expected cause for the peaks at 2.5k, 2.7k and 5.5k ?

2. Would you agree that the majority of distortion is introduced at (or before) the I/V stage ? Except maybe a little 3rd is introduced, typical of the differential stage.

3. Is it reasonable to expect a cleaner signal at the I/V? I can't know if the problem is the SPDIF source or D/A.

4. Results good/bad/average/useless ?

[SY]

The loopback looks very noisy to start with. The broadening of the fundamental on the plot is probably an artifact and suggests that you need to optimize sample rate versus number of points or the exact frequency of the test signal, as well as apodization. Get that one right before you start measuring external devices. Ritemark may be useful software for getting the measurement system in order.

[zigzagflux]

Okay, I can play around with the loopback for starters. Is the goal to get a noise floor around -120 dB ?

Can't find any info on Ritemark software here or on Google...

[zigzagflux]

How's this for a loopback? Still some distortion on the output, but the noise is much improved.

[SY]

Your noise floor will depend on resolution (# of data points and sample rate). How are those set? And how are you measuring actual levels?

[zigzagflux]

There was a number of settings in Arta, where I just followed the instructions, but some of them were not explained very well. ASIO buffer size and F/W buffer size in msec. Biggest improvement was in selecting Kaiser5 over Hann, also per Arta.

FFT length for the screenshot was 16K at 96 kHz. Averages 64. Turning the averages down did increase the noise floor a bit, but not like it used to be.

I verified my output signal is not distorted, then turned it down 2-3 dB to make sure it was running clean. Input volume turned up just before entering the red zone, but definitely before the input circuit clips. Went through the voltage calibration with a Fluke 87, but this doesn't work well in Arta. Close enough, but not really accurate.

I am a little suspicious of the sample rate. My Saffire control panel is set for 44.1 kHz, and I am able to select 96 kHz through Arta, but I don't hear the internal relays clicking in the soundcard when this is done. When you select 96 kHz through the Saffire control panel, you hear the relays clicking. However, when I start up Arta it forces the soundcard back to 44.1. So I'll to mess around with it some more.

[scott wurcer]

Originally Posted by zigzagflux

There was a number of settings in Arta, where I just followed the instructions, but some of them were not explained very well. ASIO buffer size and F/W buffer size in msec. Biggest improvement was in selecting Kaiser5 over Hann, also per Arta.

FFT length for the screenshot was 16K at 96 kHz. Averages 64. Turning the averages down did increase the noise floor a bit, but not like it used to be.

I verified my output signal is not distorted, then turned it down 2-3 dB to make sure it was running clean. Input volume turned up just before entering the red zone, but definitely before the input circuit clips. Went through the voltage calibration with a Fluke 87, but this doesn't work well in Arta. Close enough, but not really accurate.

I am a little suspicious of the sample rate. My Saffire control panel is set for 44.1 kHz, and I am able to select 96 kHz through Arta, but I don't hear the internal relays clicking in the soundcard when this is done. When you select 96 kHz through the Saffire control panel, you hear the relays clicking. However, when I start up Arta it forces the soundcard back to 44.1. So I'll to mess around with it some more.

You need to correct for the equivalent noise BW of the windowing function as well. This can be as much as several dB. High order windows being the worst (in general). When this is done your true RMS meter and sound card will agree (as long as you brick wall the input to the meter at Fs/2).

[SY]

What are the options for number of points? 16k points at 96kHz sampling is probably good enough for doing things like amplifier distortion, but if you want to get in tight for looking at low level sidebands, you'll want to use a larger number of points. 64 is a lot of averages, especially when the sample record gets longer- remember that the noise goes as the square root of number of averages, so you won't see night-and-day differences with only 16 averages. Not that more averages will hurt, it's just that it can be pretty time consuming.

Does ARTA also give you the option of Blackman-Harris apodization?

[zigzagflux]

Scott:
Sorry, but you're losing me. Although I am an engineer and passed advanced math in college with a B, I dropped out of communications systems because it was worse than Greek to me. So a lot of terminology with algorithms and noise is above me. How does a windowing function affect measuring the output voltage of a soundcard? I would think the high frequency noise present from the sampling rate would be essentially unmeasurable by a multimeter.

Sy:
I can go from 4K to 128K. Blackman-Harris 3 or 4 (whatever that means).

This is a 128K BH4 with 4 averages. I confirmed I am running at 96kHz.

At the end of the day, I want to measure my preamps/amps distortion, maybe IMD and simple speaker measurements (tweeter polarity, room correction). Eventually, I will output SPDIF to my D/A, so is the assumption this will be a perfect sine wave for measurements?

[SY]

That's pretty good- still more noise spikes than I get with a PCI card, but everything well below -100dB. More than adequate for most preamp/amp distortion measurements.

edit: Not noise spikes, harmonic components. Sorry.