TOPIC:

Hello,

While testing how well HA-DSP supports 44.1 / 48 kHz modes, I discovered an issue which looks to be introduced by the resampler. Below is a plot comparing the frequency response of HA-DSP when being operated in 48 kHz mode. The green graph is when the DSP is bypassed, the red graph is what happens when you engage the DSP, even when all the filters are being bypassed:



For your convenience, here is the red graph alone:



I have never seen those ripples in the HF area when operating HA-DSP in 96 kHz mode, so I suppose they are introduced by the resampler when upsampling to incoming wave to the DSP sampling rate.

This is quite sad, since there is a lot of music in 44.1 or 48 still.

devteam, could you please confirm whether you can reproduce this, and are there any plans to address this issue?

@ Naganov

Very odd indeed.. I wonder if it's simply a measurement artifact. Let us try to explain:

When you enable processing, it adds latency. Especially with FIR, you need to let the signal "settle" before you take a measurement at each frequency. If the sweep is too fast, it will create artifacts like you're seeing now.. On our Audio Precision, there is a setting that allows the tone to be present for "long enough" that the audio is passing through the FIR blocks (adding quite a lot of latency).
My 2cts based on this measurement (as we've seen this ) is that maybe your settling time isn't long enough to account for the latency.

Just a thought as we've never seen this issue before, but we'll still contact the engineers to confirm.

DevTeam

Thanks for your reply!

I created my signal based on FuzzMeasure's exported test signal (48 kHz / 32-bit). I took the wave file FM has created, attenuated it by 3 dB, then added 1 second of silence before the impulse (since the filter length is 1792 taps, it's processing time at 96 kHz is a couple of tens of milliseconds), and added silent right channel -- I've attached the resulting wave.

For recording, I've configured HA-DSP to 48 kHz, configured the recording interface (MOTU MicroBook IIc) also to 48 kHz, used 2 laptops on battery power to avoid creating ground loops either via power connection, or via USB grounds. Then I imported the results as "field recording" into FuzzMeasure.

I'm also getting the same ripples when I run the measurement in FuzzMeasure directly, this time using a USB isolator box to avoid creating a USB ground loop.

Considering your comments on the DSP "settle down" time, please note that I'm getting these ripples even if I set all the IIR and FIR filters to BYPASS in HA-DSP settings (but not bypassing the DSP entirely, as when the lights on HA-DSP show "xoox").

And I'm not getting these ripples when measuring in with 2 PCs, but running everything at 96 kHz. I'm not getting them when use use 48 kHz, but bypassing DSP. That's why I'm suspecting it is a SRC issue.

Would be glad to hear any suggestions on how to improve my rig. I can't offer buying AP

@ Naganov

Oops correct that we forgot that it's not happening at 96k. We'll do a chart on the Audio Precision later on this week to seek what we get...

FYI, that scaling is a bit zoomed in an extreme way though i just realized as I was wondering why it's rolling off at 10k.. ... I'm not sure how it's setup on your side for the scaling.. (i.e. that's -1dB at the bottom so that ripple would be circa 0.01dB? i.e. circa 1/10th oscillation of a 0.1dB step based on a rough estimate with a rule of thumb looking at that picture)

Yes, you are correct--this is the section of the plot in the vertical range from +0.1 dB to -1 dB, so every vertical divide is a 0.1 dB step.

The magnification is very close, I agree. This is because I'm currently working with delicate filters, where the entire filter amplitude is only 3 dB.