FPGA Central

I have a DPSK modulated signal that is sampled at 8kHz and a demodulator
that works at 7.2kHz, which is a multiple of the bit rate. The base tone is
either 1200 or 2400 Hz.

I'm looking for an effective way of demodulating the input signal. The
method I'm currently considering consists of interpolating using a FIR to
72kHz with a LPF with cutoff frequency of about 3-4kHz, then decimating
(probably just by dropping samples, but maybe with another LPF) to 7.2kHz
and pumping through the existing democulator. I have a couple of
questions:

1) Given that there's only a very small band of interest (say 1200Hz or
2400Hz +- 5%) is there any simpler way of achieving this.

2) DPSK demodulation requires detection of phase shifts. A reasonably
uniform delay wouldn't affect this, but varying phase shifts would. Is
there anything I'd need to watch to keep the phase shift information intact.

Thanks,
Alf Katz
[email protected]

Unbeliever wrote:
> I have a DPSK modulated signal that is sampled at 8kHz and a demodulator
> that works at 7.2kHz, which is a multiple of the bit rate. The base tone is
> either 1200 or 2400 Hz.
>
> I'm looking for an effective way of demodulating the input signal. The
> method I'm currently considering consists of interpolating using a FIR to
> 72kHz with a LPF with cutoff frequency of about 3-4kHz, then decimating
> (probably just by dropping samples, but maybe with another LPF) to 7.2kHz
> and pumping through the existing democulator. I have a couple of
> questions:
>
> 1) Given that there's only a very small band of interest (say 1200Hz or
> 2400Hz +- 5%) is there any simpler way of achieving this.
>
> 2) DPSK demodulation requires detection of phase shifts. A reasonably
> uniform delay wouldn't affect this, but varying phase shifts would. Is
> there anything I'd need to watch to keep the phase shift information intact.
>
> Thanks,
> Alf Katz
> [email protected]

Your interpolate by 9, decimate by 10 approach is reasonable. This can
probably be done with a single FIR lowpass filter. Look in a DSP book
under sample rate conversion.

I don't understand what you mean by "base tone".

The phase can typically vary linearly with frequency, as would be the
case for a symmetric FIR filter.

John

> Your interpolate by 9, decimate by 10 approach is reasonable. This can
> probably be done with a single FIR lowpass filter. Look in a DSP book
> under sample rate conversion.

Thanks, John, I've been looking at a few web references including a book
that's been made available on the web. Just making sure I wasn't missing
something simpler. I'll look into how to use a single FIR to cope with both
interpolation and decimation. Some of the references would suggest 2 rounds
of interpolate by 3 followed by either a single or double (/5, /2)
decimator for efficiency reasons, though I'm not yet sure I need the extra
efficiency. That bit should be easy to add later, though any major changes
can be a pain in Motorola DSP assembler.

>
> I don't understand what you mean by "base tone".
My terminology is probably bad here. I mean the frequency that is being
DPSK modulated (1200 Hz in one direction, 2400Hz in the other.

>
> The phase can typically vary linearly with frequency, as would be the
> case for a symmetric FIR filter.
>
I guess my issue will be how this affects the discontinuity where the phase
shifts by 90, 180 or 270 degrees at up to 600 Hz. My personal knowledge
base only contains empirical tools to address this issue.

Cheers,
Alf Katz
[email protected]

Unbeliever wrote:

> I have a DPSK modulated signal that is sampled at 8kHz and a demodulator
> that works at 7.2kHz, which is a multiple of the bit rate. The base tone is
> either 1200 or 2400 Hz.
>
> I'm looking for an effective way of demodulating the input signal. The
> method I'm currently considering consists of interpolating using a FIR to
> 72kHz with a LPF with cutoff frequency of about 3-4kHz, then decimating
> (probably just by dropping samples, but maybe with another LPF) to 7.2kHz
> and pumping through the existing democulator. I have a couple of
> questions:
>
> 1) Given that there's only a very small band of interest (say 1200Hz or
> 2400Hz +- 5%) is there any simpler way of achieving this.
>
> 2) DPSK demodulation requires detection of phase shifts. A reasonably
> uniform delay wouldn't affect this, but varying phase shifts would. Is
> there anything I'd need to watch to keep the phase shift information intact.
>
> Thanks,
> Alf Katz
> [email protected]
>
>
For the resampling look for information on polyphase filtering. They
give you the same mathematical effect as sampling up then sampling down,
but you only perform the computation at the down-sampled rate, so you
save clock ticks on your processor.

What is your baud rate? Your "+- 5%" implies baud rates of 30 and 60Hz
for the 1200 and 2400Hz carriers -- are they really that low?

If you use symmetrical FIR filters your delay will be uniform.

--
-------------------------------------------
Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Unbeliever wrote:
> > Your interpolate by 9, decimate by 10 approach is reasonable. This can
> > probably be done with a single FIR lowpass filter. Look in a DSP book
> > under sample rate conversion.
>
> Thanks, John, I've been looking at a few web references including a book
> that's been made available on the web. Just making sure I wasn't missing
> something simpler. I'll look into how to use a single FIR to cope with both
> interpolation and decimation. Some of the references would suggest 2 rounds
> of interpolate by 3 followed by either a single or double (/5, /2)
> decimator for efficiency reasons, though I'm not yet sure I need the extra
> efficiency. That bit should be easy to add later, though any major changes
> can be a pain in Motorola DSP assembler.
>
> >
> > I don't understand what you mean by "base tone".
> My terminology is probably bad here. I mean the frequency that is being
> DPSK modulated (1200 Hz in one direction, 2400Hz in the other.
>
> >
> > The phase can typically vary linearly with frequency, as would be the
> > case for a symmetric FIR filter.
> >
> I guess my issue will be how this affects the discontinuity where the phase
> shifts by 90, 180 or 270 degrees at up to 600 Hz. My personal knowledge
> base only contains empirical tools to address this issue.
>
> Cheers,
> Alf Katz
> [email protected]

If you use symmetric FIR filters then the group delay will be flat and
you will not have phase distortion. If you are just getting comfortable
with multirate techniques, and processor time is not an issue, then you
can do the interpolation and decimation in two steps. Once you get it
working well then go back and refine the implementation to minimize
unnecessary calculations.

Good luck,

John