Splitting audio

[geoffw64]

Hi Dan et al,

KFR rocks! I have developed a nice DSP on my Mac. I have a 7.1 configuration in my home theatre and apply a series biquad filters and other treatments to all 8 channels. All filtering is done using KFR and all samples are stored in univector and univector2d. This works and sounds great.

I filter low frequency content for all 7 main channels. I want to retain the low frequency content that is filtered out and pass it to the LFE signal (the subs do a much better job on this content). This is the same sort of bass management that any AV processor does.

Question: How should I do this with KFR? Surely it can' be as simple as subtracting the filtered samples' univector from the original samples' univector?

I am after the highest possible audio quality and have plenty of spare CPU power and RAM.

Thanks in advance,
Geoff.

[geoffw64]

Here is the filter set generated by REW for one of my channels. LS is Low Shelf, PK is Peaking, HS is High Shelf.
As you can see, the first 3 filter out the low frequencies at around 24dB below 80Hz.

Notes:0.8dB slope
Extended
Number Enabled Control Type Frequency(Hz) Gain(dB) Q Bandwidth(Hz) TargetT60(ms) Shape Slope_dB/oct Frequency2(Hz) Q2 Slope2_dB/oct
1 True Manual LS 81.20 -11.80
2 True Manual LS 83.50 -11.90
3 True Auto PK 106.5 -4.70 8.480 12.56
4 True Auto PK 127.0 12.00 1.886 67.34
5 True Auto PK 128.5 -8.80 15.453 8.32
6 True Auto PK 158.0 -16.40 1.454 108.7
7 True Auto PK 283.0 6.70 3.791 74.65
8 True Auto PK 576.0 -2.00 8.011 71.90
9 True Auto PK 785.0 6.50 2.011 390.4
10 True Auto PK 820.0 -5.70 7.363 111.4
11 True Auto PK 993.0 2.60 4.102 242.1
12 True Auto PK 1532 -4.60 1.027 1492
13 True Auto PK 2160 -3.70 5.635 383.3
14 True Auto PK 3177 3.50 1.133 2804
15 True Auto PK 4051 -4.20 4.519 896.4
16 True Auto PK 4756 -2.20 3.627 1311
17 True Auto PK 9781 -7.10 1.428 6849
18 True Manual HS 9857 11.90
19 True Auto PK 15406 0.00 1.116 13805
20 True Auto None

[geoffw64]

Hi again,

I've done some digging around and implemented the following for the 7 main channels (Left, Right Centre etc):

• Apply two 2nd order Butterworth low pass filters sequentially. Send output to LFE channel.
• Apply two 2nd order Butterworth high pass filters sequentially. Retain output to the main channel being processed.

I think this is a 24dB crossover with unity gain. Could someone wiser than me please confirm that this is right or wrong?

Thanks in advance,
Geoff.

[dancazarin]

Hi Geoff,

Your setup looks correct, and it implements a 4th-order Linkwitz-Riley crossover (two cascaded 2nd-order Butterworth filters) with unity gain. This filter provides good quality and phase alignment. Note that the LFE channel requires a gain adjustment (typically +10 dB) after summing, to match standard home theater levels, though this depends on your system’s calibration.

The code is as simple as:

double cutoff_freq = 80;
auto lp_filter = biquad_lowpass(cutoff_freq / sample_rate, 0.7071); // Q = 0.7071 for Butterworth
auto hp_filter = biquad_highpass(cutoff_freq / sample_rate, 0.7071);

Then, lp_filter must be applied twice to the original signal to extract the LFE content. Similarly, hp_filter must be applied twice to the original signal to retain the main channel content. After that, you can apply additional filters to achieve the desired frequency response.

Dan

[geoffw64]

Hi Dan,

Thanks for your very helpful response.
I created the filters using an approach in one of your examples/tests:
zpk lp_zpk = iir_lowpass(butterworth(order), frequency, sample_rate);
iir_params lp_params = to_sos(lp_zpk);
const iir_filter lp_filter_1(lp_params);
Of course II use iir_highpass for the other side of the crossover.
Filters constructed at dsp start up time.
Filters applied twice in succession at playback time you described.

As you mentioned, the bass needs to be attenuated. Some content has significant bass in all 7 channels. I'm looking at you Jim Kerr (Simple Minds) on the New Gold Dream Bluray audio! :)
I'll raise a separate discussion item on ways to manage this smartly, rather than just attenuate it all by several dB as I am currently doing.