Hi guys. I'm working on a project for my lab class (Digital Signal Processing). Being an audio guy, I of course wanted to do something w/audio and stay away from image processing. So I decided auto-eq'in (like the DEQ2496) would be a fun, interesting idea.
Does anyone have some nice links to theory/algorithms for how to go about implementing auto-eq? I'm planning on bringing my DEQ2496 to provide pink noise to our DSP chip we are programing (TMS320C54). I have a wide variety of libraries avaliable to me (FFT,IFFT,autocorr. . .) so please give me some links!
Or perhaps impart your massive amount of digital knowledge on me.
I've been googling but nothing seems like a "gold mine" of good advice, I want some info I can trust.
Thanks.
Does anyone have some nice links to theory/algorithms for how to go about implementing auto-eq? I'm planning on bringing my DEQ2496 to provide pink noise to our DSP chip we are programing (TMS320C54). I have a wide variety of libraries avaliable to me (FFT,IFFT,autocorr. . .) so please give me some links!
Or perhaps impart your massive amount of digital knowledge on me.
I've been googling but nothing seems like a "gold mine" of good advice, I want some info I can trust.
Thanks.
You are entering a difficult field
First, you need to read up on papers in this subject. JASC might be a good start.
Then a good measurement means getting the IR or frequency/phas eresponse. Pink noise measurements are not good for this. Use either WinMLS or another acoustic impulse response measurement program.
You should consider if you want a single sample, or measure the response over a volume (typical listening positions), then doing some kind of averaging.
The inverse response is a problem as well. A straight inversion of frequency response results in an overly bright (and excessive clipping) sound. My best bet would be modelling the woofer as a 2nd/4th order highpass system, and not correcting it.
with regards
Knut Inge
First, you need to read up on papers in this subject. JASC might be a good start.
Then a good measurement means getting the IR or frequency/phas eresponse. Pink noise measurements are not good for this. Use either WinMLS or another acoustic impulse response measurement program.
You should consider if you want a single sample, or measure the response over a volume (typical listening positions), then doing some kind of averaging.
The inverse response is a problem as well. A straight inversion of frequency response results in an overly bright (and excessive clipping) sound. My best bet would be modelling the woofer as a 2nd/4th order highpass system, and not correcting it.
with regards
Knut Inge
As Knut has correctly pointed out, doing auto/non-auto room EQ properly is rather a difficult field of endeavour.
The other approach to room eq is to use tones stepped in minute amounts (look up numerically controlled oscillators (NCO) and direct digital synthesis (DDS)) and examining how the room/speakers respond to different frequencies... this is simple from a signal processing point of view and overcomes to some extent the problems inherent with calculating impulse responses...
As an alternative, perhaps create a DSP system that makes a 3 way crossover entirely in the digital domain?
This of course would involve conversion of SPDIF to IIS format data (requires PLLs so you may want to do this in hardware initially) and then the extraction of the serial audio data into parallel data for left and right channels, then the three filters (6 in total for stereo). You can then extend this system to more frequency bands...
Alternatively maybe examine and vary the phase relationships of the L and R signals to show how surround sound 'images' and sound stages are created by varying the phase of the two signals... if you are really keen, maybe re-create 5.1??
Perhaps look into audio compression formats and algorithms (Wavelets/DCT etc)? WiFi?
From past experience - pick a topic you are interested in and start simple, get that going, then add bits to it... that way when the deadline is upon you, you can just hand in the previous version, or squeeze the next one in...
Good luck! It is an interesting, challenging and often frustrating topic to delve into. Just measuring stuff to make sure you have it right is a pain in the ...
apollyon25
The other approach to room eq is to use tones stepped in minute amounts (look up numerically controlled oscillators (NCO) and direct digital synthesis (DDS)) and examining how the room/speakers respond to different frequencies... this is simple from a signal processing point of view and overcomes to some extent the problems inherent with calculating impulse responses...
As an alternative, perhaps create a DSP system that makes a 3 way crossover entirely in the digital domain?
This of course would involve conversion of SPDIF to IIS format data (requires PLLs so you may want to do this in hardware initially) and then the extraction of the serial audio data into parallel data for left and right channels, then the three filters (6 in total for stereo). You can then extend this system to more frequency bands...
Alternatively maybe examine and vary the phase relationships of the L and R signals to show how surround sound 'images' and sound stages are created by varying the phase of the two signals... if you are really keen, maybe re-create 5.1??
Perhaps look into audio compression formats and algorithms (Wavelets/DCT etc)? WiFi?
From past experience - pick a topic you are interested in and start simple, get that going, then add bits to it... that way when the deadline is upon you, you can just hand in the previous version, or squeeze the next one in...
Good luck! It is an interesting, challenging and often frustrating topic to delve into. Just measuring stuff to make sure you have it right is a pain in the ...
apollyon25
If you are in a hurry or with a deadline limit for delivering your work, it may be not enough time to sucessfully develop the algorithms (because measuring, checking, error-checking, measuring again, and finally write code...).
If you like audio and do not want to do image processing may be you could consider to build a multieffects unit (chorus, flanger, delay, reverb, eq, etc.).
Good luck.
If you like audio and do not want to do image processing may be you could consider to build a multieffects unit (chorus, flanger, delay, reverb, eq, etc.).
Good luck.
Ok so the project is well underway and SEEMS to be going well. I have a quick question though. Perhaps I should have asked this in the planning stage, but why do RTA equalizers, such as the DEQ2496, flatten the frequency response of the loudspeakers with respect to pink noise? I understand that pink noise has equal energy per octave, but why send pink noise through the speakers and then equalize the response of the speakers so that the mic input is receiving pink noise (or at least close to it)? Couldn't white noise, or any noise for that matter, be used to do the same thing? Is it because upon breaking the frequency spectrum into octaves it's easier to compare the power spectrum since with pink noise there's equal energy per octave?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.