Spectran beta 4 is released
Alberto di Bene
Mon, 31 Jul 2000 11:42:29 +0200
SPECTRAN beta 4, build 110 has been just released.
Beta 4 is a major enhancement with respect to beta 3.
Full duplex, real time audio processing has been added,
with denoising, bandpassing and CW peaking functions.
Here are some of the new features :
- Display of the frequency where the mouse cursor is hovering on.
- Possibility to switch off the visual AGC (not the audio AGC, but
that AGC that in beta 3 was always applied when computing the
color of the spectrogram). It has been brought to our attention that
with strong QRN this AGC had the side effect of 'chopping' otherwise
quite readable QRSS signals, making difficult to discriminate
between dots and lines.
- More consistent timing when computing the overlap factor, leading
to a more constant speed of the spectrogram.
- Semiautomatic procedure for the setting of the Windows mixer.
The Windows mixer is the most non-standard feature of Win9x,
a real nightmare (thanks Bill...). Lacking tight specifications,
every soundcard manufacturer has felt free to implement the mixer
functions in the way they pleased them more. Hopefully the procedure
we have devised to set the mixer should take care of this.
- Denoiser : the denoiser is of the Widrow-Hoff type, i.e. a FIR filter
whose h coefficients are adjusted in real time with the LMS (Least Mean
Square) algorithm, making the filter to converge towards a dynamic
solution that minimizes the so-called AWGN noise. We have devised a
variant of the standard method, where the leaking factor applied to the
h coefficients has been made proportional to the mean amplitude of the
- Bandpass : from the low-cut and high-cut limits set with the mouse a
rectangular windows is defined, its IFFT is computed, then a Hamming
window is applied to it, and finally a forward FFT is computed, leading
to a 'well-shaped' filtering window that minimizes the Gibbs phenomenon.
This window is applied to the FFT of the audio signal, which is then
brought again in the time domain with an IFFT, ready for further processing.
You can see a simulation done with Matlab of the final result at this
URL : http://www.weaksignals.com/images/bandpass.gif
The irregularities seen on the stop band are due to the simulation of
rounding the coefficients to 5 significant digits.
- CW Peaking : this function is simply an IIR resonator, with two zeros
on the real axis at -1 and 1, and two complex conjugate poles at a radius
of 0.995. The angle of the poles is computed in real time, when the frequency
of the desired peak is set with the mouse. This kind of resonator has been
choosen by personal judgement, listening to its effect on CW signals. We
feel that its shape is much more adequate for CW than a 'brick-wall' filter,
even if the stop-band rejection could be better. But you can always use the
CW peak filter and the bandpass filter in cascade, if you so prefer. For those
wanting to see the interesting shape of this filter, there is another
Matlab simulation here : http://www.weaksignals.com/images/cwpeak.gif
A word of warning : the bandpass of this filter is roughly 30 Hz, so if used
with very fast CW it could smooth a bit too much the keying component.
These more or less are the major enhancements of beta 4. Only a few final words
on how to use them. For the mixer setting, Spectran will recognize the first
execution of beta 4 on a given machine, and will automatically start the setup
procedure. Just follow the simple instructions on the screen, and you will not
be bothered again.
To set the filters, a filter definition panel has been added. If you briefly tap
on either Shift key, the panel will appear. Another tap on the Shift key, and it
will disappear. When it is visible, the left mouse button will set the lower limit
of the bandpass, the right mouse button the upper limit. To set the CW peak frequency,
use the combination Ctrl-Left Mouse Button.
To actually switch the filters on and off the audio processing chain, use the red leds
on the left side of the main Spectran panel. If you want to use them all
simultaneously, you need at least a 166 MHz Pentium if you limit the sampling
frequency to 11025 Hz. For 22050 Hz a 266 MHz Pentium should do.
The 'PassThru' led, when choosen, simply activates the rerouting of the input
signal, unprocessed, to the output. In this mode the signal does not incur in the
delay of about one second between input and output, due to the buffering and
processing. It is intended to make tuning easier (which it isn't with the delay...)
This should be all. Sorry for the long message, but a bit of explanations were in
order. One last thing : we need your feedback. We 'think' our mixer setting
procedure should take care of all the possible variations around (it did in our
few test cases), but there is always a case you did not anticipate... so please
tell us how it works on your computer with your soundcard. Thanks.
Spectran beta 4 is freely available here : http:/www.weaksignals.com