Difference between calculated pre-gains and maximum of individual gain is 0.06dB for my use case - pretty small difference.
I agree to your point about probability of distortion since I won't listen to 0dB impulse or step function.
Pre-Equalization gain for parametric equalization (PEQ)
- Thread starter jed1
- Start date
Difference between calculated pre-gains and maximum of individual gain is 0.06dB for my use case - pretty small difference.
I agree to your point about probability of distortion since I won't listen to 0dB impulse or step function.
Brantome
Moderator
See this and related posts
PEQ talk
for wiim team... no news ? time flies quickly... ;-) is it so difficult to add bands to this peq? (see a choice of distribution of these in L or R? Besides, we never follow a simple balance function) hardware dsp power problems have these been revealed? roon's certification may be out of...
forum.wiimhome.com
Brantome
Moderator
Yeah, good idea. Sounds like a feature request ticket
That's why I started this topic, but I didn't manage to request a ticket. I wrote to them from the app.
3.1 dB headroom, according to REW. To be on the safe side you would have to set the volume limit to 94%.
Are you sure that this +7.9 dB peak at 76 Hz is really helpful when moving your head just slightly from the center position?
If you really need it that much then I don't understand why you not upvote the feature request created by @jed1 .
Sorry, I'm still new, I'm posting on mobile, I accidentally missed the like. 76 Hz is right, it's really a lot. Thanks.
Max volume, in the first place. Available resolution, if you go over the top (at least in theory).
80% would give you 12 dB of headroom for positive gain, which is quite a lot, probably much more than you'll need.
With 80% I have no problem getting enough SPL from my HiFi chain.
I don’t understand “Available resolution, if you go over the top (at least in theory)”…
Can you please elaborate a bit?
Turho76,
I wrote a Python script and calculated the impulse response.
With the current PEQ setting, you need -0.93dB pre-EQ gain to avoid clipping. If you are fine with removing filters #7-#10, you need -0.00044884373667138284dB (i.e. 0dB).
Below is a Python script - it works in Windows, Mac OS and etc. Forum does not allow attaching a Python script so I dump as below.
import numpy as np
from scipy.signal import lfilter
# reference: https://webaudio.github.io/Audio-EQ-Cookbook/audio-eq-cookbook.html
def parametric_eq(bands, fs, pregain):
# Initialize impulse response
impulse = np.zeros(1000)
impulse[0] = 1.0*pow(10,pregain/20)
for band in bands:
filter_type, center_freq, gain, Q = band
A = pow(10,gain/40)
omega = 2*np.pi*center_freq / fs
cos_omega = np.cos(omega)
sin_omega = np.sin(omega)
alpha = sin_omega / (2*Q)
# Define filter types
if filter_type == 'low_shelf':
b0 = A*((A+1) - (A-1)*cos_omega + 2*np.sqrt(A)*alpha)
b1 = 2*A*((A-1) - (A+1)*cos_omega)
b2 = A*((A+1) - (A-1)*cos_omega - 2*np.sqrt(A)*alpha)
a0 = (A+1) + (A-1)*cos_omega + 2*np.sqrt(A)*alpha
a1 = -2*((A-1) + (A+1)*cos_omega)
a2 = (A+1) + (A-1)*cos_omega - 2*np.sqrt(A)*alpha
elif filter_type == 'high_shelf':
b0 = A*((A+1) + (A-1)*cos_omega + 2*np.sqrt(A)*alpha)
b1 = -2*A*((A-1) + (A+1)*cos_omega)
b2 = A*((A+1) + (A-1)*cos_omega - 2*np.sqrt(A)*alpha)
a0 = (A+1) - (A-1)*cos_omega + 2*np.sqrt(A)*alpha
a1 = 2*((A-1) - (A+1)*cos_omega)
a2 = (A+1) - (A-1)*cos_omega - 2*np.sqrt(A)*alpha
elif filter_type == 'peaking':
b0 = 1 + alpha*A
b1 = -2*cos_omega
b2 = 1 - alpha*A
a0 = 1 + alpha/A
a1 = -2*cos_omega
a2 = 1 - alpha/A
elif filter_type == 'low_pass':
b0 = (1 - cos_omega)/2
b1 = 1 - cos_omega
b2 = (1 - cos_omega)/2
a0 = 1 + alpha
a1 = -2*cos_omega
a2 = 1 - alpha
elif filter_type == 'high_pass':
b0 = (1 + cos_omega)/2
b1 = -(1 + cos_omega)
b2 = (1 + cos_omega)/2
a0 = 1 + alpha
a1 = -2*cos_omega
a2 = 1 - alpha
elif filter_type == 'band_pass':
b0 = Q*alpha
b1 = 0
b2 = -Q*alpha
a0 = 1 + alpha
a1 = -2*cos_omega
a2 = 1 - alpha
elif filter_type == 'all_pass':
b0 = 1 - alpha
b1 = -2*cos_omega
b2 = 1 + alpha
a0 = 1 + alpha
a1 = -2*cos_omega
a2 = 1 - alpha
elif filter_type == 'notch':
b0 = 1
b1 = -2*cos_omega
b2 = 1
a0 = 1 + alpha
a1 = -2*cos_omega
a2 = 1 - alpha
else:
raise ValueError('Invalid filter type. Choose either "low_shelf", "high_shelf", or "peaking".')
# Normalize filter coefficients
b = [b0 / a0, b1 / a0, b2 / a0]
a = [a0 / a0, a1 / a0, a2 / a0]
# Calculate impulse response for this band and add to total response
response = lfilter(b, a, impulse)
impulse = response
print([center_freq,min(impulse),max(impulse)])
return impulse
bands = [('peaking', 28.0, 2.0, 1.00),
('peaking', 53.5, -6.4, 1.75),
('peaking', 76.0, 7.9, 1.65),
('peaking', 100.0, -3.0, 0.65),
('peaking', 120.0, -3.6, 1.65),
('peaking', 155.0, 5.0, 1.25),
('peaking', 2000.0, 0.0, 0.25),
('peaking', 4000.0, 0.0, 0.25),
('peaking', 8000.0, 0.0, 0.25),
('peaking',22000.0, 1.3, 0.20)]
response = parametric_eq(bands, 96000,0)
20*np.log10(max(response))
bands_trunc = [('peaking', 28.0, 2.0, 1.00),
('peaking', 53.5, -6.4, 1.75),
('peaking', 76.0, 7.9, 1.65),
('peaking', 100.0, -3.0, 0.65),
('peaking', 120.0, -3.6, 1.65),
('peaking', 155.0, 5.0, 1.25)]
response_trunc = parametric_eq(bands_trunc, 96000,0)
20*np.log10(max(response))
I assume WiiM apply IIR filters (biquad) sequentially. If that assumption holds, as long as each IIR filter step does not clip, the final filtered signal is not clipped.
slartibartfast
Major Contributor
- Joined
- Apr 18, 2024
- Messages
- 430
Where would I see this calculation in REW? I must have missed it.
Brantome
Moderator
slartibartfast
Major Contributor
- Joined
- Apr 18, 2024
- Messages
- 430
Thanks. I must have seen that every time I used REW and completely ignored it
Impulse response through IIR filters will simulate how DSP works internally. As long as there is no clipping at each step, there will be no clipping at the end. Main benefit is to use a higher gain than REW (it seems to use maximum spectral response for headroom).