I'm bumping a thread I started last year, as I've still not been able to find the solution.
I've been using some old (2007) Faust code, originally developed by Albert Graef, and modified slightly by me, in one of my Max creations. I compiled the Faust code into a Max external. I've recently been trying to update my Max application to 64 bit, so I recompiled the Faust code (using the online IDE) to create a 64 bit Max external. That worked fine (with one minor tweak that Stephane Letz helped me with), but the external is not behaving correctly. Giuseppe Silvi helped me to update the calls to the previous libraries that were used (math.lib and music.lib) to the current stdfaust.lib in the code listed below, but the behavior is still not correct. I have two videos that illustrate the difference in behavior between old and new versions, which hopefully gives clues to where the problem lies - you'll see differences in the 4th and 5th audio outputs: Working 32 bit-version: https://defectiverecords.com/temp/kcs202debug-Max7-32bit.mp4 <https://defectiverecords.com/temp/kcs202debug-Max7-32bit.mp4> Non-working 64 bit version: https://defectiverecords.com/temp/kcs202debug-Max8-64bit.mp4 <https://defectiverecords.com/temp/kcs202debug-Max8-64bit.mp4> The code is very well documented as you can see below, but sadly my Faust skills are not up to debugging this (not to mention that I don't really understand how it works in the first place!). I'm happy to post the original code if at all helpful. My best guess is that a function in one of the original libraries is not equivalent to that used in the new stdfaust.lib? Thank you for any tips in advance! Dan -- Dan Nigrin / Defective Records / https://defectiverecords.com <https://defectiverecords.com/> Fauxmo: a Fizmo editor / Cyclic, M185 & Klee Sequencers / MC-4, MC-202 and DSX Hacks /* The Kansas City Standard (KCS) uses frequency shift keying to encode binary digits at 300 bps as an audio signal. This Faust module provides a KCS signal analyzer for the Casio FA-3 tape interface used with the early Casio Basic pocket computers from the 1980s, such as the Casio PB-100 and the FX-730P. You can find more info about the hardware of these PDA precursors at http://www.pisi.com.pl/piotr433 <http://www.pisi.com.pl/piotr433>. */ /* This got quite a bit more involved than I first made it out to be. Maybe that's also the reason why I couldn't find any ready-made solution on the net that would work with the noisy output signal produced by my Casio PB-100 calculator. Thanks are due to Jos Horsmeier for pointing out that autocorrelation might do the trick, which it actually did. :) */ // MODIFIED BY DAN NIGRIN December 2007 import("stdfaust.lib"); //import("math.lib"); //import("music.lib"); /* Here are a few threshold constants you might have to tinker with. The default values seem to work ok with the FA-3 interface, but YMMV. */ K0 = 0.1; // absolute threshold of noise gate K1 = 0.3; // rectifier threshold (relative to max amplitude) // DAN NIGRIN CHANGE // K2 = 0.8; // rising edges of autocorrelation signal // take out the constant, and make a variable K2 = hslider("K2", 0.8, 0.5, 1.0, 0.01); // DAN NIGRIN CHANGE // K3 = 0.3; // falling edges of autocorrelation signal // take out the constant, and make a variable K3 = hslider("K3", 0.3, 0.1, 0.6, 0.01); /* Maximum size of various delay lines; this must always be a power of 2 and larger than the window size n (see below). */ N = 1024; /* Frequency of the 1-bit base tone (which is also the frequency of the lead-in tone). For standard KCS this is 2400 Hz, but the frequencies of the Casio FA-3 tape interface are slightly shifted because it uses a 32768Hz xtal as a reference, resulting in 32768Hz/14=2340Hz as logic 1. In any case, 8 cycles of this wave represent a 1 bit, while a 0 bit consists of 4 cycles of half the frequency. The length of one symbol thus corresponds to SR*8/freq samples which is also used as the window size for the autocorrelation analysis. */ freq = hslider("freq", 2340, 1000, 5000, 1); k = ma.SR/freq : ceil : int; // period length (samples) n = ma.SR*4/freq : ceil : int; // symbol length (samples) - MODIFIED BY DAN NIGRIN FROM SR*8 /* Input gain in dB. Use this to make up for a weak input signal. Changes to this value will also trigger a recalibration of the signal envelop (see below). */ gain = ba.db2linear(hslider("gain", 0, -96, 96, 0.1)); /* 1 pole LP used to estimate the current signal strength. This unit is used to trigger processing as soon as the signal rises above a certain absolute threshold. */ env = abs : *(1-g) : + ~ *(g) with { t = 0.1; // attack/release time in seconds g = exp(-1/(ma.SR*t)); // corresponding gain factor }; /* Determine the maximum amplitude used to compute the threshold for rectifying the signal. This gets recalibrated automagically when t=1 (which happens in the main program as soon as the signal rises above a certain threshold or the input gain is modified). */ maxenv(t) = *(1-t) : max ~ *(1-t); /* Rectify a signal using a given threshold, and turn the rectified signal into a train of "trigger" pulses. */ rect(m,x) = x>=m; trigger(x) = x>x'; /* Integrate a signal over a window with given length d. */ integrate(d,x) = x-de.delay(N,d,x) : + ~ _; /* Sample-and-hold. Repeats last nonzero value of the input signal. */ samplehold(x) = rwtable(2, 0, x==0, x, 0); /* Clock signal. Repeats last nonzero value of the input signal every d samples. */ clock(d,x) = rwtable(N, 0, x==0, x, (int : %(int(d)) : *(x==0)) ~ +(1)); /* Signal cleanup. This removes "nearby" double pulses (less than d samples away from each other) in the output signal. Only the second pulse of each such pair is kept. NOTE: Since a Faust program processes signals "on-line", i.e., it cannot look into the future, this unit delays the input signal by d samples. */ cleanup(d,x) = (1-v)*de.delay(N, d, x) with { u = integrate(d, abs(x))>1; v = trigger(1-u); }; /* The main program. The input is the mono signal x0 to be analyzed. The main output is the y1 signal. A single-sample +1 pulse means a binary 1, a -1 pulse a binary 0. In between the symbol pulses the y1 signal is 0. */ /* For illustrative and debugging purposes we also output the preamplified input signal, the gate signal (useful to sense when a transmission starts and stops), the rectified input signal and the rectified autocorrelation signal. Essentially, the latter is the rectified autocorrelation of the rectified input signal which is then sampled at the symbol rate to construct the output signal. Some complications arise, though, because the detected 0 and 1 bit pulses are not fully in sync and the symbol length rounded to samples is not fully accurate either. This is taken care of by a final "cleanup" phase in which any resulting spurious double pulses are removed from the output signal. */ process(x0) = attach(x1, e : hbargraph("maximum", 0, 1)), g1, r1, s1, y1 with { /* Apply input gain. */ x = x0*gain; /* Noise gate. */ g = env(x)>K0; /* Rectified input signal, with pulses of length k for each period of the input signal. This is done by detecting the rising edges during the first half of the signal period. */ e = maxenv((g>g')|(gain!=gain'), x); r = integrate(k, trigger(rect(K1*e,x)*g)) : min(1); /* Autocorrelation of the rectified input signal. Note that the rectified signal looks more like 1-1-1-1-... for 1-bits, whereas 0-bits look more like 1-0-1-0-..., where each 0 or 1 represents one base frequency period of k samples here. Hence the 1-bit signal's autocorrelation for a delta of k samples will be close to 1, whereas the 0-bit signal produces an autocorrelation close to 0. */ a = integrate(n, r*de.delay(N, k, r))/n; /* Lead bits corresponding to rising and falling edges of the autocorrelation signal (1 = 1 bit, -1 = 0 bit). */ b = g*(trigger(a>=K2)-trigger(a<=K3)); /* Clock signal (pulses at symbol rate, triggered by lead bits). */ c = g*(clock(n, b>0)-clock(n, b<0)); /* Sample-and-hold of the lead bits, yielding the rectified autocorrelation signal. */ s = g*samplehold(b); /* The preliminary output signal, which is obtained from the rectified autocorrelation signal by sampling at the pulses given by the clock signal. In effect, this repeats the lead bits at the symbol rate. */ y = (c>0)*(s>0)-(c<0)*(s<0); /* Finally clean up the output signal to remove any spurious double pulses which are less than n2 = half the period away from each other. Note that this delays the signal by n2 samples, so we apply a corresponding delay to the other output signals as well. */ n2 = int(n/2); y1 = cleanup(n2, y); x1 = de.delay(N, n2, x); g1 = de.delay(N, n2, g); r1 = de.delay(N, n2, r); s1 = de.delay(N, n2, s); }; -- Dan Nigrin - Defective Records - https://defectiverecords.com Fauxmo: a Fizmo editor / Cyclic, M185 & Klee Sequencers / DSX, MC-4 & MC-202 Hack
_______________________________________________ Faudiostream-users mailing list Faudiostream-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/faudiostream-users
