applying esp-dsp IIR biquad to pipeline or element ?

[shabtronic]

that's my PC code - VST2 - it's all doubles!

u need floats or whatever the DSP lib takes - I assumed float - but I've not looked at it at all

I'll post my full EQ later - once I've tidied it up - it's not actually mine - came from a blog - in turn came from RBJ's cookbook. But I did add the Z-transform stuff

[benbiles]

did a test bypassing DSP int16 --> float --> int16 and its stereo again.

you are correct , ESP32 has no problem with that conversion.

so something strange happening to the samples in the DSP library making the samples mono and a small buzzing !

will post here if I can find out why.

[benbiles]

fixed the buzzing ( was samples out of range +1 -1 float )

I (443) EQUALIZER: Impulse response of IIR filter with F=0.250000, qFactor=4.000000
I (453) view: Data min[2] = -1.535849, Data max[0] = 1.975187
________________________________________________________________
0xxx |
1xxxxx |
2xxxxxxx |
3xxxxxxxxxxx |
4xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
5xxxxxxxx |
6xxxxxx |
7xxxx |
8xx |
9x |
0123456789012345678901234567890123456789012345678901234567890123


I divided the DSP processed float values in half

// half dsp outpt values sould be less than +/-1 ****************
float devideit = 0.5;
for ( int z = 0; z < len; z++ )
{
FloatDspBufB[z] = FloatDspBufB[z] * devideit; // half
}

tried different values for the dsps_biquad_gen_lpf_f32

float freq = 0.25;
float qFactor = 0.5;

the it works without buzzing.

Now just to work out why its going into mono ! maybe the delay line is messing up things.

[benbiles]

Awsome if you want to share your EQ code later !
would love to see how it's working.

[shabtronic]

Here you go - here's my old filter code - it's for PC Vst2. But it gives a great a insight into the coefficient calculations and the Freq response Plotting code.

I wouldn't use this - because - you know the IDF and ADF so far for me have been superb - it's fast and quick to get stuff up and running. I havn't looked at the DSP lib, but these espressif folk know what they are doing with the esp32 FPU - so it's gonna a fair amount of work to blindly optimize that code.

But have fun anyhow!

https://github.com/shabtronic/Audio-Filters

[benbiles]

Wow, thanks for that code! Will definately be useful as a reference.

I'm slowly getting there. Narrowing down simple things. the coefficient I was manually supplying to the DSP lib were in double and not float ( 8 byte rather than 4 byte)

I still have the 2 channels turning to mono as soon as I process the float samples in the array.
I aslo ran the biquad code in C and it did the same. It probely just used a few more cpu cycles.

Just need to find out what's changing to make everything mono. The only thing I can think of is the I2S is left justified and processing the samples with the biquad maths is changing to right justified.

I should probebly send a bunch of &FFFFFFFF followed by & 00000000 as samples and take a look at the binary values before and after the maths and check justification of the bits.

[shabtronic]

cool stuff!

I noticed on your previous code your doing this:

Code: Select all

for ( int i = 0; i < len; i++ )
{
// do this properly with ESP-DSP maths ?
FloatDspBuf = ((float)DspBuf) / (float)32768;
}

I know it's pseudo code - but it should be /2 because len is in bytes

Code: Select all

for ( int i = 0; i < len/2; i++ )
{
// do this properly with ESP-DSP maths ?
FloatDspBuf[i] = ((float)DspBuf[i]) / (float)32768;
}

The raw audio data from the i2s stream is interleaved, with array element 0=left,1=right,2=left,3=right e.t.c. (or the otherway around - depends on the driver e.t.c.)

Havn't look at the dsp stuff tho - so I can't help you - it's half the "fun" working all this stuff out

have fun!

[benbiles]

Ah, good to know len is in bytes! Had wrongly presumed it was just an int.

I'll try changing that now. I assume the len value is doubled when I have stereo streams. Will take a better look at that.

[shabtronic]

this would be the code if your I2S stream is 16bit:


elsewhere in code

Code: Select all

BiQuad MyFilter;
MyFilter.CalcCoeffs(4,000,1,1,44100,-1);
float MyFloatBuf[1024];

in DSP Function:

Code: Select all

// convert to floats
for (int a=0;a<len/2;a++)
	MyFloatBuf[a]=(float)((short*)Buf)[a]/32768;

// Process Left/Right samples    /4 this time because we are processing Left/Right
for (int a=0;a<len/4;a++)
	MyFilter.Process(MyFloatBuf[a*2+0],MyFloatBuf[a*2+1]);

// Convert back to stream
for (int a=0;a<len/2;a++)
	((short*)Buf)[a]=MyFloatBuf[a]*32768;

Havn't tested that, just wrote it for clarity, probably some errors - but you get the idea!

Have fun

[benbiles]

Dps fullily working now?

I'm just processing all the samples in the
Buffer in one go because I'm applying the same effect to both left and right channel.

I change to interleaved in case I need stereo fx.

It turns out my mono problem is a hardware fault or possibly a firmware that needs updating.

I convinced myself stereo pass through was working at one point but I had only enabled the direct line input to output feature of the codec.

If I just flash basic pass through example either mic or line, channel 1 inputs on ADC (mic) or channel 2 inputs on ADC (line aux connector) I get a mix of both channels.

I have another lyrat 4.3 board in the post so I'll check that soon as I arrives.

I flashed the esp32 to latest bootloader firmware but did'nt help.

I the lyrat 4.3 manual it shows 3 other bin files at diffent address and so far I have no idea what they are. I better start a new post for that.

So anyway, a massive thanks for you help / interest.

The original equalizer.c element input / output samples buffer is set to 100 length , seams like an odd number

So far I have to say I'm impressed with the lyrat Esp-adf / esp32 framework.

OK going to post some firmware questions before I assume the board is defective but its looking like it might be.