Audio cut-off really at 1.6kHz?

[pixel]

Is this really the case? Because I thought I'd have heard higher frequencies and that's really not the kind of bandwidth where one can understand digitised speech, or can one?

[Noizer]

In my ears it’s not an absolute cut off, the high and middle frequencies are there. Try to connect some audio equipment to rise the highs. I think therefore no need for modifications on the sacred motherboard

[Mike]

The cut-off frequency of a low-pass filter is defined as the frequency, where the attenuation is at 3 dB, i.e. 70% of the maximum signal amplitude. To a first approximation, you have 100% of the signal below cut-off, and then the attenuation increases with 20 dB per decade.

In the VIC-20 audio circuitry, two RC filters define one cut-off frequency at 1.6 kHz (as you wrote) and another at 16 kHz. Between 1.6 kHz and 16 kHz, the attenuation increases with 20 dB/decade, above that with 40 dB/decade.

So, in principle you can still output a, say, 20 kHz sine wave - but it will be very quiet.

[pixel]

In my ears it’s not an absolute cut off, the high and middle frequencies are there. Try to connect some audio equipment to rise the highs. I think therefore no need for modifications on the sacred motherboard

Just would love to know what's going on really. Could help others to improve VICE and me to think up better audio code.

[pixel]

The cut-off frequency of a low-pass filter is defined as the frequency, where the attenuation is at 3 dB, i.e. 70% of the maximum signal amplitude. To a first approximation, you have 100% of the signal below cut-off, and then the attenuation increases with 20 dB per decade.

In the VIC-20 audio circuitry, two RC filters define one cut-off frequency at 1.6 kHz (as you wrote) and another at 16 kHz. Between 1.6 kHz and 16 kHz, the attenuation increases with 20 dB/decade, above that with 40 dB/decade.

So, in principle you can still output a, say, 20 kHz sine wave - but it will be very quiet.

So I guess Fourier transform it is? Cannot see the standard LPF algorithm do such thing.

[Mike]

So I guess Fourier transform it is? Cannot see the standard LPF algorithm do such thing.

It's perfectly possible to do frequency filtering in the time domain.

Simple example: y(i) = 0.1 * y(i-1) + 0.9 * x(i), with x(i): input and y(i): output. The cut-off frequency depends on the sample rate and the coefficients.

Keywords: IIR filter, step-invariant Z-transform.

[pixel]

So I guess Fourier transform it is? Cannot see the standard LPF algorithm do such thing.

It's perfectly possible to do frequency filtering in the time domain.

Simple example: y(i) = 0.1 * y(i-1) + 0.9 * x(i), with x(i): input and y(i): output. The cut-off frequency depends on the sample rate and the coefficients.

Keywords: IIR filter, step-invariant Z-transform.

Yeah. Right. My math teacher probably just died by laughing too hard...

[pixel]

Someone patched the low-pass filter in VICE and will probably commit it to the repository, soon. What remains is the terrible aliasing noise – but the same guy is interested in adding a resampler to get around it.

[groepaz]

That someone was Leandro, who is also the main driving force behind reSID fixing these days

The patch is merged now, have fun

[Noizer]

In my ears it’s not an absolute cut off...

Just would love to know what's going on really. Could help others to improve VICE and me to think up better audio code.

Mike is right, it‘s very possible the output samples > 16KHz with the standard vic20 hardware.
Therefore there is no need to cut off frequency above the 16KHz when preparing samples for feeding, as seeing on spectral view of Amen Brother. Of course, this rises the entropy of a sample, making eventually compression less efficient.
Bye for now

[Noizer]

So I guess Fourier transform it is? Cannot see the standard LPF algorithm do such thing.

It's perfectly possible to do frequency filtering in the time domain.

Simple example: y(i) = 0.1 * y(i-1) + 0.9 * x(i), with x(i): input and y(i): output. The cut-off frequency depends on the sample rate and the coefficients.

Keywords: IIR filter, step-invariant Z-transform.

Do you have some FFT (fast fourier tr.) algorithms for 65xx cpu available?

[pixel]

Do you have some FFT (fast fourier tr.) algorithms for 65xx cpu available?

Nah. Was thinking for VICE but it only needs a resampler now as the filter is emulated perfectly (I assume).

[Mike]

The main issue is not with any filtering (or lack thereof) on the output side, i.e. when converting digital/sampled signals to analog output.

Main issue is the input side or when anything is resampled, keyword: Nyquist frequency. If there are any significant signal components at or above the Nyquist frequency (which is 1/2 the sampling frequency) in the input signal, they're folded/mirrored back into the baseband frequency range and produce alias frequencies. They're then what is heard as shrieking noise. Once they've been introduced, nothing can be done to eliminate them!

Now, the fun fact about Sigma-Delta modulation is: as long as you keep within the allowed dynamic range, its amplitude response is flat (i.e. = 1, or 0 dB attenuation) up to the Nyquist frequency and it merely inherits the characteristic of any filter attached to its output - see:

H. Inose, Y. Yasuda, J. Murakami, "A Telemetering System by Code Manipulation - ΔΣ Modulation" , Ire Trans on Space Electronics and Telemetry, Sep. 1962, pp. 204-209

[Noizer]

All perhaps interesting stuff. I thought instead something more in direction of Digital information solution or algorithms

[Mike]

All perhaps interesting stuff. [...]

We are talking about the basics of signal processing here. You would better get them right before throwing anything like a FFT against a rather simple problem like low-pass filtering audio signals.