Hi all,

I'm new to pure data but prepared to learn a little bit if I can get it to solve my problem! Basically, I need to process all audio out of my computer so that my AV receiver will be able to separate the bass to direct to my subwoofer. To do this all I need to do is filter off everything below 100Hz and send half to the left channel and invert the other half to send to the right channel. The Dolby Pro-Logic int he receiver will then decode this and direct the bass to the rear speaker output, where my subwoofer is plugged in!

My big question is whether pure data is powerful enough to do this. It would need to perform a low pass filter and high pass filter of my live audio output, which doesn't sound too taxing. Hardware wise I have a Creative XtremeGamer soundcard, onboard sound, and a quad-core pc.

If anyone could give me a few pointers that would be great!

Thanks,

James

Sure. There are several kinds of filters in iemlib (included in Pd-extended) that will probably do the job well enough. If you want a higher quality crossover, you'd likely have to design an FIR filter to use with [FIR~], but luckily brick-wall FIR filters are well documented, if you even want to go that route.

Also, you'll have to use something like Jack or Soundflower to get the audio into Pd if you want to play them from another app.

Ahh Jack might remove my current need to have a cable between line out and line in on one of my sound cards!

I've jus used lop~ and hip~ at the moment as I've never used this program before - will different filters do a better job?

Yes. [hip~] and [lop~] are first order filters and don't have a very steep transition. I personally don't find them to even be good for what they are, as they don't do anything if the cutoff is above about 7000 Hz.

You might want to try the high-order Butterworth filters in iemlib, as these have flat pass bands and sharper transitions. They're called [lp10_butt~] or [hp10_butt~], with the 10 being the order (I believe 10 is the highest). The only issue with them is that they distort the phase, but with such a low cut-off frequency it might not be so perceptible, as most of the distortion is around the transition band. Usually it's the transients that are affected by the filter's phase response, since sharp transients rely on all the frequencies lining up. But they also have a lot of high-end content, which won't be so affected.

If it is an issue, then that's when you want to go the FIR route.

Actually, now that I think about it, the Bessel filters might be a better choice. They have a nicer phase response than Butterworth filters, but not as steep a transition. But with an order of 10, it's still probably steep enough.

OK I've changed it to those ones, but I'll have to wait until the morning to see how it sounds. I'll let you know how it sounds! Thanks for the help

Just out of curiosity, what's the purpose of inverting one of the signals?

I've been running a two speaker setup for a while, but fancied a bit more punch, so I bought a passive subwoofer. My amp doesn't have a special subwoofer output so it's a bit tricky to separate it, and I can only pass signals between my PC and the amp using a 3.5mm (stereo headphone type jack) to phono cable.

What Pd is doing for me is inverting everything <100Hz on one side so that the bass signal is out of phase between left and right. Then I can get my bass separation to the sub in one of two ways:

1. wire the sub between the positive terminals of L + R speakers (L+ to Sub+, R+ to Sub-). Only out of phase signals will cause a potential difference, so the bass signal will pass through the subwoofer because I made it out of phase.

2. Read up on how Dolby Surround works - basically the rear channel is encoded in a very similar way to how I've encoded the bass. By plugging my subwoofer into the rear output of my amp (which has only stereo inputs but dolby pro-logic and surround outputs) I get the bass signal through the sub.

Both of these worked in practise, but number 1 was more successful - it sounded brilliant, the sub giving it some punch and all the higher frequencies through the smaller left and right speakers. I'm going to be fiddling with setting for a while yet to optimise it, but really happy at the moment.

One thing I need to do is learn how to use Jack, because I currently have to use a wire from out->in on one of the soundcards to make it an input in Pd. That should cut a bit of noise and latency out, although I can't hear any noise and the latency is 50ms, but it might help.

*edit: I have no idea if this idea has been incorporated into commercial sound systems or not - does anyone know? After reading how Dolby worked it seemed the obvious solution.

Ah, I see. It's an encoding/decoding thing. In that case, if you're not doing so already, make sure you also run the non-inverted channel through filters to ensure that both channels have the same phase distortion. That will make sure that they are equally out-of-phase.

If I have some time later, I'll post up something using a linear-phase FIR filter. Since this is a phase-dependent system, it will probably sound a little better. And I'm a bit curious.

Yep - everything is through filters, here's the code: http://s7.postimage.org/i9ifx52uz/Pdbass.png
The values aren't rigorously derived, just what sounds best, and there are other EQ things going on outside the program (Bass/treble knobs on amp, soundcard driver etc).

First time I've ever come across Pd (wish I'd known about it sooner!) so any tips are appreciated. Can Pd output Bode plots for each channel?

It can, but you have to patch it to. In Help -> Browser -> PureData -> 3.audio.examples -> H10.measurements.pd is an example of how to find the frequency and phase response of an arbitrary filter, but it's not logarithmic.

I also wrote an abstractions called [spectrum.mmb~] that...wait for it...plots the spectrum of a signal. It can be either linear or logarithmic. So you could just pump [noise~] through a filter and plot it using [spectrum.mmb~] to get an idea of it's frequency response. It won't give you phase, though. It's part of my library:

https://github.com/dotmmb/mmb

Hey je305,

Here's a version using linear-phase FIR filters. I don't have the system to test it, so I'm just relying on plots.

http://www.pdpatchrepo.info/hurleur/subbass.mmb.pd

Hey je305,

Here's a version using linear-phase FIR filters. I don't have the system to test it, so I'm just relying on plots.

http://www.pdpatchrepo.info/hurleur/subbass.mmb.pd

Thanks a lot, I've give it a test. I've just been comparing lop~ and lp10_bess~ and the Bessel filters do a noticably better job

Wow. There is so much more clarity, especially in the middle ranges.
There seems to be a very strong sense of stereo - I'm not sure if this is just because stereo effects were being lost/muddied in the previous system or if it's because the two sides are being treated differently, I'll do a bit of experimentation with stereo and mono inputs.

Thanks very much, this is turning out to be pretty interesting!

*I think it's more of an unbalanced sense of stereo, hard to put my finger on exactly.

**This program requires the bass to be centred perfectly to work ideally, I'm going to have a quick think about adjusting it so that the bass can be imbalanced and still work (I think this was the source of the sense of stereo imbalance somehow or other). No doubt the program will become less elegant to accommodate this, which is a shame, but it should make it more robust. I have a plot of the panning of frequencies being created by a friend at the moment, I'll post it up if suggests bass imbalance.

http://s15.postimage.org/rg4s3dz8b/linescatter.png This shows the bass is a little over to the left.

Here is an attempt at balancing it

http://www.pdpatchrepo.info/hurleur/subbasbalanceds.mmb.pd

Typically crossovers should be Linkwitz-Riley filters (two Butterworth filters in series) to ensure that frequency response is flat when summed.

Just took a quick look at how Linkwitz-Riley filters work, and it sounds like the better option. I just tried using a shorter length FIR and running noise through it, and there's definitely a notch at the cutoff point. Thanks for pointing that out, acreil.

Compared the balanced-FIR patch with a Linkwitz-Riley setup using order 10 Butterworths in series and the FIR one does sound better. May well have done something wrong though.

10th order seems like a little much. That would give you a quite long impulse response, and a lot of phase shift. You don't need an extremely steep slope.

Ah ok, 2nd order sounds a lot better.