Audio Equalizer Filter design

Started by olsene01 April 19, 2011
I have been attempting to design a 15 band audio equalizer using Simulink. My design right now is as follows (please let me know if there are any significant design flaws here):

-Audio signal goes into preamp.
-Preamp output is split into 15 branches, 1 for each filter.
-Each branch contains a BPF, followed by a gain block.
-The 15 branches are summed, and that is my output.

The output sounds decent, except for low frequencies seem quite distorted, so I started to do some trouble shooting. Checked out the phase for each filter... they all appear to be approximately linear within the passband, and what happens to the phase after the passband is irrelevant, right?

Next, I noticed something a little more troubling with the low frequency filters don't nearly meet the passband edge frequency characteristics I specify, since the sampling frequency must be 44100-Hz. I guess my main question is how should I design the filters such that their magnitude specifications are properly met, using such a high sampling frequency? Is there a better design method to use for audio equalizers?

Also, can someone explain Q values a little better for me, or point me to a good reference? From what I've read, a constant Q value throughout the equalizer is the best design, but trying to implement various constant Q values made my current design sound worse.
Olsen-

Audio equalizers are a complex subject. I know people who have spent chunks of their careers developing AEs with
better performance, with very low power consumption (suitable for embedded systems), that equalize based on an
arbitrary room response, etc.

So what you're doing is not easy. Suggest that you do lots of online research and find other academic projects. In
particular, you might search on parametric filters.

With that said, what type of filters are you using? If you want high quality low frequency filters, you will need to
use IIR filters. Also lower frequency IIR filters will require high numerical precision to be stable.

-Jeff

> I have been attempting to design a 15 band audio equalizer using Simulink. My design right now is as follows (please
> let me know if there are any significant design flaws here):
>
> -Audio signal goes into preamp.
> -Preamp output is split into 15 branches, 1 for each filter.
> -Each branch contains a BPF, followed by a gain block.
> -The 15 branches are summed, and that is my output.
>
> The output sounds decent, except for low frequencies seem quite distorted, so I started to do some trouble shooting.
> Checked out the phase for each filter... they all appear to be approximately linear within the passband, and what
> happens to the phase after the passband is irrelevant, right?
>
> Next, I noticed something a little more troubling with the low frequency filters don't nearly meet the passband edge
> frequency characteristics I specify, since the sampling frequency must be 44100-Hz. I guess my main question is how
> should I design the filters such that their magnitude specifications are properly met, using such a high sampling
> frequency? Is there a better design method to use for audio equalizers?
>
> Also, can someone explain Q values a little better for me, or point me to a good reference? From what I've read, a
> constant Q value throughout the equalizer is the best design, but trying to implement various constant Q values made
> my current design sound worse.








I found a few years ago a technique for making high performance bell curves for the implementation of parametric or graphic EQ. The technique uses an all pass filter whose output is added or subtracted with the source signal.

 

I don't recall details, but used the method on code for a Freescale 56311,       Even at low frequencies and high Q, it accurately matched Matlab simulations.

 

For references, search for Dana Massie and Sanjit Mitra (coverage n a text text book).

 

I believe the method is also good from the standpoints of coefficient sensitivity, noise, and simplicity of parameter variation.

 

To use for a graphic, you will cascade the band sections, not parallel them.

 

Chris Moore











__._,_.___