Software clock recovery for biphase demodulation

* Does anyone know of source code for decoding Biphase/Manchester
encoding from analog signal?

I am looking for source code for a biphase decoder. The application is
extracting SMPTE/EBU time code data from analog audio signals (recorded
by professional audio/video equipment). I found source code that looks
at signal crossing an adjustable threshold (by Maarten de Boer)

    ftp://www.iua.upf.es/pub/mdeboer/projects/SMPTE/

However, this is not robust for real-world signals.

I assume I need a clock recovery algorithm, then sample the signal
between clock edges.

* Where can I find a clock-recovery software algorithm, or source code?

If there is none, here is what I would try doing:
- create a narrow band-pass filter 2.4 kHz
- recover clock by running the data signal through the filter and
obtain close-to-sine-wave around 2.4kHz
- detect zero-crossing of the recovered clock
- sample or integrate data signal between the zero-crossings to detect
positive or negative symbol.

* Is this the right way to go?
* Which band-pass filter should I use: Butterworth, Chebyshev,
Elliptic, other?
* The clock frequency shifts so I might need a PLL, or an adaptable
filter. Any advice there?

More information on the encoding is at:
    http://www.philrees.co.uk/articles/timecode.htm

Thanks,

Laza

Does your module need to support variable speed SMPTE time code or only fixed
speed?  Variable speed is usually only obtained with analog recorders with the
time code recorded (striped) to an analog track.  Then when the analog recorder
is played back slower or faster (varispeeded), the time code signal changes as
well.

If you don't need to support variable speed, I've found that zero-crossing
detection works quite well, as long as sufficient hysteresis is used.  Some
pre-filtering might not be a bad idea, but I found it wasn't strictly necessary
in my application.

Checking for errors in the sync word, illegal time values, or time that is way
out of whack with the previous frame(s) is a good way to add robustness as well,
and can all be easily done in a post-processing step.

"Laza" <zoranlazarevic@yahoo.com> wrote in message
news:1107804001.755081.11210@c13g2000cwb.googlegroups.com...
> * Does anyone know of source code for decoding Biphase/Manchester
> encoding from analog signal?
>
> I am looking for source code for a biphase decoder. The application is
> extracting SMPTE/EBU time code data from analog audio signals (recorded
> by professional audio/video equipment). I found source code that looks
> at signal crossing an adjustable threshold (by Maarten de Boer)
>
>     ftp://www.iua.upf.es/pub/mdeboer/projects/SMPTE/
>
> However, this is not robust for real-world signals.
>
> I assume I need a clock recovery algorithm, then sample the signal
> between clock edges.
>
> * Where can I find a clock-recovery software algorithm, or source code?
>
> If there is none, here is what I would try doing:
> - create a narrow band-pass filter 2.4 kHz
> - recover clock by running the data signal through the filter and
> obtain close-to-sine-wave around 2.4kHz
> - detect zero-crossing of the recovered clock
> - sample or integrate data signal between the zero-crossings to detect
> positive or negative symbol.
>
> * Is this the right way to go?
> * Which band-pass filter should I use: Butterworth, Chebyshev,
> Elliptic, other?
> * The clock frequency shifts so I might need a PLL, or an adaptable
> filter. Any advice there?
>
> More information on the encoding is at:
>     http://www.philrees.co.uk/articles/timecode.htm
>
> Thanks,
>
> Laza
>

No, I do not need variable speed time code. However, my signal is
coming from a minidisc recording in MP3 format. I tried running it
through Maarten de Boer's code (threshold = 50% of max) and it did not
decode any correct time codes.

BTW: since all is done in software (Winamp/Windows Media Player), even
when speeding up the playback, the same number of samples are still
being processed. That is equivalent to constant speed time code.
However, the speed probably shifts a little and might not be 100%
stable due to dubbing from analog tapes to digital format.

Thanks for advice.

I've never looked at time code after MP3 compression/decompression.  Have you
looked at the decompressed waveform with a wave editor or oscilloscope?  Does it
look reasonable?  I'm not at all familiar with Maarten de Boer's code, but it
well may that the underlying mechanism is sound but there is a simple bug/issue
with the implementation.

I'd be interested in knowing what effect MP3 has on SMPTE time code!  Back when
I was working on it, there was no MP3 (at least in common use) and since I was
usually dealing with analog recordings, I was concerned with issues like noise,
wow/flutter, frequency response, cross-talk, etc..

"Laza" <zoranlazarevic@yahoo.com> wrote in message
news:1107807998.557671.265220@c13g2000cwb.googlegroups.com...
> No, I do not need variable speed time code. However, my signal is
> coming from a minidisc recording in MP3 format. I tried running it
> through Maarten de Boer's code (threshold = 50% of max) and it did not
> decode any correct time codes.
>
> BTW: since all is done in software (Winamp/Windows Media Player), even
> when speeding up the playback, the same number of samples are still
> being processed. That is equivalent to constant speed time code.
> However, the speed probably shifts a little and might not be 100%
> stable due to dubbing from analog tapes to digital format.
>
> Thanks for advice.

Laza wrote:
> * Does anyone know of source code for decoding Biphase/Manchester
> encoding from analog signal?
>
> I am looking for source code for a biphase decoder. The application
is
> extracting SMPTE/EBU time code data from analog audio signals
(recorded
> by professional audio/video equipment). I found source code that
looks
> at signal crossing an adjustable threshold (by Maarten de Boer)
>
>     ftp://www.iua.upf.es/pub/mdeboer/projects/SMPTE/
>
> However, this is not robust for real-world signals.
>
> I assume I need a clock recovery algorithm, then sample the signal
> between clock edges.
>
> * Where can I find a clock-recovery software algorithm, or source
code?
>
> If there is none, here is what I would try doing:
> - create a narrow band-pass filter 2.4 kHz
> - recover clock by running the data signal through the filter and
> obtain close-to-sine-wave around 2.4kHz
> - detect zero-crossing of the recovered clock
> - sample or integrate data signal between the zero-crossings to
detect
> positive or negative symbol.
>
> * Is this the right way to go?
> * Which band-pass filter should I use: Butterworth, Chebyshev,
> Elliptic, other?
> * The clock frequency shifts so I might need a PLL, or an adaptable
> filter. Any advice there?
>
> More information on the encoding is at:
>     http://www.philrees.co.uk/articles/timecode.htm
>
> Thanks,
>
> Laza

I come from a communications background, so some of these comments may
not apply to audio. To recover clock from a PSK waveform, one can
subject the signal to a nonlinearity to collapse the modulation into a
single tone at the bit rate or a multiple of it. Example nonlinearities
include absolute value and half-bit delay and multiply. The
nonlinearity is sometimes preceded by a filter and always followed by
one. I have had good luck with 2nd order Butterworth bandpass filters
after the nonlinearity. After the BPF, one can trigger on alternate
zero crossings to obtain sampling instants. If the clock frequency
varies, the zero crossing will track, but the BPF bandwidth must be
wide enough to accomodate the variation.

Sometimes a PLL is used after the BPF to lock onto the recovered clock
and provide a "flywheel" effect where the trigger instants continue
even in the absence of zero crossings due to transitionless patterns
and signal fades.

Hope that helps some.

John

Laza wrote:

>No, I do not need variable speed time code. However, my signal is
>coming from a minidisc recording in MP3 format. I tried running it
>through Maarten de Boer's code (threshold = 50% of max) and it did not
>decode any correct time codes.
>
>BTW: since all is done in software (Winamp/Windows Media Player), even
>when speeding up the playback, the same number of samples are still
>being processed. That is equivalent to constant speed time code.
>However, the speed probably shifts a little and might not be 100%
>stable due to dubbing from analog tapes to digital format.
>
>Thanks for advice.
>  
>
The ear is largely insensitive to phase above about 2kHz. Music 
compression, therefore, saves bits by making no attempt to preserve the 
phase. I doubt any phase modulation technique will pass through MP3 
compression/decompression.

Regards,
Steve

Minidisc's (normally) do not use MP3 compression.

Minidiscs normally use ATRAC compression which is similar perceptual
compression but not the same as MP3.

It would be hard to predeict what MP3 or ATRAC would do to the SMPTE
signal.  Your first step should be to examine the playback signal and
see if it is even possible to recover the data.  

Mark

John, thanks a lot. I think I will try the 2nd order Butterworth
bandpass filter.

I looked at the signal in Scilab software (free Matlab clone). Instead
of having square waves, the signal is very rounded.

The signal is bi-polar (either +1 or -1). Shifting by 1/2 bit and
multiplying is not enough because that generates two clock pulses for
"1" and one clock pulse for "0". Applying absolute value yields
constant signal (+1)


Steve,

If the phase is lost in MP3 compression, how would that affect
compressing a square wave train? I assume there would be a delay or an
advance introduced randomly. Right?

Rounding of the signal shouldn't make it unrecoverable, as long as

What bitrate of MP3 were you using?  I did a quick informal experiment on a
snippet of SMPTE LTC I found on the web and found that with sufficient bitrate
(say > 128kbps) the signal looked pretty good even after MP3
compression/decompression.  But I was just looking at the waveform, not trying
to actually decode it.

To address another point, is it really MP3 or is it the ATRAC usually used on
minidisc?  I suppose you can use minidisc as a generic storage mechanism, and
hence use any type of file/compression you like, but if it is recorded as a
standard audio device, it will be ATRAC (at least AFIAK).

"Laza" <zoranlazarevic@yahoo.com> wrote in message
news:1107879397.409906.123360@g14g2000cwa.googlegroups.com...
> John, thanks a lot. I think I will try the 2nd order Butterworth
> bandpass filter.
>
> I looked at the signal in Scilab software (free Matlab clone). Instead
> of having square waves, the signal is very rounded.
>
> The signal is bi-polar (either +1 or -1). Shifting by 1/2 bit and
> multiplying is not enough because that generates two clock pulses for
> "1" and one clock pulse for "0". Applying absolute value yields
> constant signal (+1)
>
>
> Steve,
>
> If the phase is lost in MP3 compression, how would that affect
> compressing a square wave train? I assume there would be a delay or an
> advance introduced randomly. Right?
>

I will check this. I actually have the files on my hard disk in MP3
format, and I am told they come from a minidisc.