2x4 minidsp

[pos]

http://www.minidsp.com/applications/...itz-lab-lxmini

[ivica]

I would suggest to reread the linked page (onwards) more carefully

Hi POS,

Knowing You as DSP expert in design digital filters, do You have any experiments in order to phase linearized IIR filter by implementing cascade of IIR all-pass filters, such as explained in the:

"EFFICIENT GROUP DELAY EQUALIZATION OF DISCRETE-TIME IIR FILTERS" by M.F.Quélhas,A.Petraglia,M.R.Petraglia
http://www.eurasip.org/Proceedings/E...ers/cr1483.pdf

AN EFFICIENT METHOD OF GROUP DELAY EQUALIZATION FOR DIGITAL IIR FILTERS by Piotr Okoniewski, Jacek Piskorowski
http://www.degruyter.com/view/j/mms....-2013-0034.xml

or something here:
http://www.nt.tuwien.ac.at/fileadmin.../diss_Lang.pdf


regards
ivica

[pos]

Hi Ivica
Well, I am no DSP expert.
Phase linearization using multiple all-pass filters is a proven technique that is for example used by PSI Audio down to the low midrange in their studio monitors, using only analog components (CPR).
Compared to FIR the implied delay will be the same for a given phase correction, but I suspect it will generate much more quantization noise (multiple recursive filters).
And of course a given target will be much more difficult to approach with this technique than with FIR.

[ivica]

Hi Ivica
Well, I am no DSP expert.
..............
Compared to FIR the implied delay will be the same for a given phase correction, but I suspect it will generate much more quantization noise (multiple recursive filters).

Hi POS,

I can agree with You that quantization noise can rise (multiple recursive filters) , but using today 32-bit floating-point DSP ( such as ADSP21469 ) I think that such noise 'contribution' would be under the expectable level ( of -100dB), so may be You can improve Your nice work about phase correction, allowing to use much less filter taps (not to mention down and up sampling etc).

Regards
Ivica

[pos]

That would be interesting to add such a feature in rePhase indeed, but I don't know if all targets could be reached close enough that way.
I have a mixed biquad/FIR mode in the work that could make use of this technique. I don't have a lot of time to work on this these days though...

Quantization noise with IIR filters is real though, and needs special care when EQ has to be applied down low.
For example miniDSP typically specializes some EQ banks for LF purpose, with better precision to reduce quantization noise.

Here is an example comparing the DCX2496, the openDRC using IIR EQ (first firmware), and the openDRC using FIR EQ, with the exact same minimum-phase EQ :


Uppers traces (indistinguishable) are amplitude responses.
Lower traces are noise measurements (quantization noise from the EQ + noise floor of the measurement system and units at test)

As you can see the openDRC IIR implementation was lacking precision for LF. Now they have somewhat "fixed" the problem in the openDRC by increasing the precision of the first IIR EQs slots (so you have to make sure you use those first slots for LF purpose, and only use the remaining ones for >100Hz EQs...).

Unfortunately the noise floor of that measurement is too high to really show differences between the openDRC FIR and the DCX2496, but they are also real :
Using headphones one can clearly hear noise when the sweap passes through the EQ point, whereas the openDRC in FIR mode is totally silent.
This is easy to hear on a sweap, even if it is very low in level, because the noise is fullrange and there is no masking effect.
This would probably not be a problem with music signal, but adding multiple EQs (or all-passes) would increase the audibility...

[ivica]

That would be interesting to add such a feature in rePhase indeed, but I don't know if all targets could be reached close enough that way.
I have a mixed biquad/FIR mode in the work that could make use of this technique. I don't have a lot of time to work on this these days though...

Quantization noise with IIR filters is real though, and needs special care when EQ has to be applied down low.
For example miniDSP typically specializes some EQ banks for LF purpose, with better precision to reduce quantization noise.

Here is an example comparing the DCX2496, the openDRC using IIR EQ (first firmware), and the openDRC using FIR EQ, with the exact same minimum-phase EQ :


Uppers traces (indistinguishable) are amplitude responses.
Lower traces are noise measurements (quantization noise from the EQ + noise floor of the measurement system and units at test)

As you can see the openDRC IIR implementation was lacking precision for LF. Now they have somewhat "fixed" the problem in the openDRC by increasing the precision of the first IIR EQs slots (so you have to make sure you use those first slots for LF purpose, and only use the remaining ones for >100Hz EQs...).

Unfortunately the noise floor of that measurement is too high to really show differences between the openDRC FIR and the DCX2496, but they are also real :
Using headphones one can clearly hear noise when the sweap passes through the EQ point, whereas the openDRC in FIR mode is totally silent.
This is easy to hear on a sweap, even if it is very low in level, because the noise is fullrange and there is no masking effect.
This would probably not be a problem with music signal, but adding multiple EQs (or all-passes) would increase the audibility...

Hi POS,

Thanks for the data about the noise level of EQ using DCX2496, OpenDRC FIR and openDRC IIR.
As I have expected DCX2496, using 32-bit floating point DSP ( ADSP21065 DSP engine) even using IIR EQ has almost the same noise level as openDRC FIR EQ ( fs~ 20Hz), even I would expect about 20dB better results for DCX2496 then presented, but may be there is some other source of the noise.

If I have understood correctly, on DCX2496 there is audible noise that can be heard using headphones, and while applying sinusoidal signal around 20Hz.
Does such noise start to reduce while reducing the depth of the applied notch around 20Hz (it would be expect-able to diminish with almost flat EQ applied)?
May be such nose can be reduced by applying a kind of dither applying to the feed-back path of the bi-quad EQ filter, if You decide to realize EQ IIR filter.

Any how, I believe that Your rePhase great work can be applied to the miniSharc DSP using IIR phase compensation in order to reduce number of the FIR taps, and computation power.

Reagrds
ivica

[pos]

I will try to redo some measurements with a better noise floor and different situations (varying Q, gain and frequency...) but I think my soundcard AD converter (and most probably analog stage) will be the limiting factor here.

I also have a EV DX46 crossover I could measure (48bit fixed), but the digital input does not work anymore (I have no idea why )

[ivica]

I will try to redo some measurements with a better noise floor and different situations (varying Q, gain and frequency...) but I think my soundcard AD converter (and most probably analog stage) will be the limiting factor here.

I also have a EV DX46 crossover I could measure (48bit fixed), but the digital input does not work anymore (I have no idea why )

HI POS,

Waiting to see the data...

Regards
ivica

[4313B]

I just thought it might be interesting to see what someone else is doing with the miniDSP.