To PD Gurus:

In my PD project I need to insert a simple 3 bands eq over 120 audio channels.

Big trouble is I'm using an old AMD Athlon processor computer and I need the best EQ for saving CPU load. I don't need extremely "quality" EQ. Simple 3 bands EQ and low CPU load.

What do you think is the best option?
Some patch or example?

Thanks a lot in advance.

120 audio channels? Whoah.

Yeah! I know it sounds incredible. But, I'm designing a "personal monitoring system" to use in stage during Live Music Shows, and I need around 120 channels. This is the reason because I need some "light" eq for my cpu.

Thanks in advance.

I think there are two options for less CPU load.
[vcf~] and [bp~] are two object of the main pd.
The [vcf~] needs more cpu but takes the frequency as an audio signal, which may thus change continuously in time without any clicks.
If you have constant values for the frequencies and Q, and you don't want to change them while audio input playing, then is better to use the [bp~] filter. You can check the help file of this object.

Thanks a lot Geonnik.
I'll take a look over this objects.

Any more ideas will be welcome.

Regards.

iemlib has loads of different filters, don't know what the CPU usage is like.

http://pure-data.svn.sourceforge.net/viewvc/pure-data/trunk/externals/iemlib/

If you really need minimum CPU, you might be best off writing a custom external and tweaking it out with CPU-saving features like denormal-bashing and SIMD stuff.

http://www.musicdsp.org/files/denormal.pdf has some awesome tips for denormal handling.

a similar problem to this popped up a while ago, and there was a great, simple solution:

don't use 120 eq's. just use 1 !

you have a lowpass filter, a bandpass, and a high pass.

then you use three [*~ ] objects on each signal to vary how much goes through the low, mid and high.

here's a good 3 band eq that you could modify:

http://www.pdpatchrepo.info/hurleur/dj-eq.pd

WHOOAAH!
Simple and great solution!!!
This sound very interesting but at this moment I can't understand exactly how to use one eq for all 120 channels. I let's go to learn something more about *~ and the dj-eq example you sent.

Remember I need EQ per channel. Using *~ I can't understand how I can do it. I think it will be useful for do a eq mix of all 120 channels. Where is my mistake?

Thank you very much. Your help is amazing for me!

@hardoff said:

a similar problem to this popped up a while ago, and there was a great, simple solution:

don't use 120 eq's. just use 1 !

you have a lowpass filter, a bandpass, and a high pass.

then you use three [*~ ] objects on each signal to vary how much goes through the low, mid and high.

[*~ ] just multiplies the signal by a number or fraction. so, if you multiply by 0.5, the volume of that signal will be half.

then, a basic eq would have stuff like this:

a) lowpass

[lop~ 500]

midpass

[hip~ 500]
|
[lop~ 2000]

c) highpass

[hip~ 2000]

for ease of drawing here, let's use a send/receive pair to show what happens to the signal in a simple one channel eq:

[inlet~]
|
[s~ my-signal]

[r~ my-signal]
|
| [0 ....-> 1]
| |
[*~ ]
|
[lop~ 500]
|
[throw~ eqd-signal]

there's the lowpass section. we just multiply the signal by a fraction between 0 and 1 to decide how much of the signal goes through the lowpass section.

here's the mid eq:

[r~ my-signal]
|
| [0 ....-> 1]
| |
[*~ ]
|
[hip~ 500]
|
[lop~ 2000]
|
[throw~ eqd-signal]

again, we just multiply the signal's amplitude to decide how much will go through the mid-eq section.

and finally the high end:

[r~ my-signal]
|
| [0 ....-> 1]
| |
[*~ ]
|
[hip~ 2000]
|
[throw~ eqd-signal]

then...we just catch~ all the signals at the end, and output to our speakers, or headphones or whatever:

[catch~ eqd-signal]
|
[dac~]

so, if you were to multiply that into 120 subpatches, every subpatch would contain 3 amplitude multiplications [*~ ] , and then 3 sets of filters.

however, because the 3 filter sections will be the same in every subpatch, you can just multiply the amplitudes (very cheap on cpu) and then output all 120 signals to one group of filters before going to the output.

if you look at the eq i posted as a patch, the aplitude multiplication occurs AFTER the filters. so it would be like:

[inlet~ ]
|
[lop~]
|
[*~ ]
|
[outlet~]

but, you can switch the order and have the same result:

[inlet~ ]
|
[*~ ]
|
[lop~ ]
|
[outlet~]

Amazing!

Thanks a lot!

Now, I understand it.

Hardoff, you are a PD God!

Thanks again.

Thanks a lot for your help in Puredata Forum.

I just tried your patch suggestion. It's working but I think somethig is wrong.

I routed the 120 signals to 1 EQ subpatch using *~ but I can not find the way to modify the EQ for every channel. When I modidy EQ I'm aplying the same EQ in all channels. Obviously this is not desired. I'm becoming crazy!

How can I do it?

Thanks a lot in advance.

Just a question to clarify:

You have 120 input signals.
How many output signals do you need?

What kind of EQ do you need?
Just low/mid/high levels with fixed frequencies? (simple EQ)
Or parametric EQ with everything changeable for each input signal?

If you just need simple EQ, you can get away with this I think, as hardoff suggested:

[inlet~]
|
+----+-----+
| | |
[*~] [*~] [*~] these adjust low/mid/high levels for this input signal
| | |
above part for each input signal
below part for each output signal
| | |
[lop~] [bp~] [hip~] do the EQ for this output signal
| | |
+-----+-------+
|
[outlet~]

If you need parametric EQ for each channel, there's no getting away from needing 120x EQs, as far as I can tell...

Well, I attach an example of my work. 16 channels with simple eq.
I don't know if this is the best way for "cpu savings".
Or maybe I can use eq patch in other way better.

Any suggestions will be welcome. Thanks again.

http://www.pdpatchrepo.info/hurleur/16ch-eqexample.pd

> I don't know if this is the best way for "cpu savings".

They are always on so it uses a lot of CPU. You should arrange for the EQ to only switch on when you move a control away from a flat response.

I can't understand how can I do it.

You could use a loadbang and messages to set the faders to a default value. The EQ abstraction that Hardoff provided already has a switch. If you compare the value of the faders with their default using [==] then the EQ can be activated as soon as someone moves a fader. You may need to wrap this so that the faders are outside the DSP. An EQ-OFF function that sets the faders back to their defaults will switch off the EQ abstraction.

In practice only a few of the channels will have active EQs, so you save a lot of CPU by only switching them on as needed.

In hardware digital mixing desks (like the Yamaha, HHB, Roland etc) you will see they have a limited number of EQ blocks divided amonst all the channels. Once you've used them all up, no more. 120 channles is quite a lot and it's unlikely more than 1/4 of all the channels would need an EQ. An audio routing system that inserts EQs as needed from a limited pool might be best.

See if you can work out how to do the automatic switching first.

a,

ok, here's how i was suggesting to do it. see attached 8 channel version.

http://www.pdpatchrepo.info/hurleur/8ch-eqexample.pd

a little off topic to hardoff

how did you make the [pd eq]s have the attached slider gui like that?

right click on an empty part of the canvas and select 'graph on parent', and then adjust the dimensions to fit the sliders.

and then for each slider, click 'init' so that the value is initialized when the patch is loaded.

I needed an eq that could modify 4 channels with the same group of faders for a four channel work, so I modded hardoff's. Thought I'd repost out of respect/ in the hope it could be useful to others.

Put it in a folder with your patch and make a new object called "eqsection".

http://www.pdpatchrepo.info/hurleur/eqsection.pd