Different measurements LADSPA / stm32

[JMF11]

Hello,

For my speakers, I have now two implementations:
- one on ecasound + LADSPA
- one on Stm32 microcontroller, same but without at the moment the delay between channels (about 0.05ms)

I took time this WE to check the frequency response of the new stm32 implementation of the filters and PEQ. I measured for both implementations: all speaker, woofer alone, tweeter alone, all speaker but with tweeter inversed. Speaker/micro in same position, same amp, same volume...

My results are OK from let's say 150hz to 20kH, but look strange to me between 30 and 150 Hz. See graphs below. If there is a good consistency above 150Hz, in the below range I have significant differences.

In a same implementation, the measurements of the full speaker and of the woofer are very consistent in that range, so it does not looks like something not possible to repeat. This seems related to my implementations. It looks like a frequency shift of something.

I use 16 bits 48k samples in the stm32 option. For the ecasound LADSPA version, it is 44.1k

Normally in that area, there is large PEQ at 50 Hz, 7dB. Crossover is at about 700 Hz.

For both implementations, I have checked the compliance of the configuration to a miniDSP config: perform the calculations and output in a file, compare the freq response of the file in Holmimpulse to a direct measurement on a miniDSP.

I used a basic mic from the audissey set-up of a Marantz AVR, when my previous micro capsule failed in the WE. the measurements were made at about 30 -50 cm from the drivers.

Below first figure: blue = LADSPA, red = stm32
second figure: blue = woofer LADSPA, red = woofer stm32, green = full speaker stm32 (you canc heck the both stm32 mesures are consistent in low end range, the same comparison for the LADSPA measurements shows the same consistency in the low range => repeatable measurements.

Would you see some possible explanation, or direction to investigate?

Kind regards,

JMF

 

[CharlieLaub]

What are those nulls around 30-60 Hz??? Are you making an 'in room' measurement? I suspect the answer is yes. If so, you are measuring the speaker+room response, and if you move the mic that will change.

I suggest that you make two woofer measurements (one with each config, e.g. stm32 or LADSPA+ecasound) with the mic about 1cm from the center of the woofer dustcap. This is called a 'nearfield' measurement, and will only contain the woofer's response because the room response will be comparatively small and doesn't influence the woofer output as seen by the mic at that position. If those two measurements look different, then start worrying if the two ways you are doing DSP are resulting in some differences.

 

[JMF11]

Thanks Charlie,

Yes, it is in room measurement. I had however not changed the positions of the speakers and mic, to be able to compare. However I will do the measurement you propose at 1 cm of the woofer. I'm a bit puzzled about how to set up the thing:

The mic is Audyssey mic Full Screen Image | Audioholics

The speakers are LXmini, so the woofer fires to the top. Would it be better to have body of the mic horizontal or vertical from the top to the bottom?. Maybe I will try both.

Could what I see come from precision of calculations at low frequency? ARM library uses 32 bits floats, when I believe LADSPA uses doubles.

Kind regards,

JMF

 

[JMF11]

Thanks Charlie, spot on (as always !).

The nearfield measurement, implemented as you suggested shows the same frequency response for the LADSPA and for the Stm32/Arm implementation.

Ouf (as we say in France).

I have still an "inverted" impulse response. Maybe this is related to a small glitch that I have when I start to play my measurement signal with my Stm32 implementation (potentially a residual bug).

Now, I still have to adapt your mTap (I'm really slow).

Kind regards,

JMF