Is it difficult to do FM demodulation digitally?

On May 5, 8:43&#4294967295;pm, Tauno Voipio <tauno.voi...@notused.fi.invalid>
wrote:
> On 5.5.11 9:09 , David Kirkby wrote:
>
> > On May 5, 4:52 pm, Vladimir Vassilevsky<nos...@nowhere.com> &#4294967295;wrote:
>
> >> I am not familiar with TS-590S architecture details; however from what
> >> you wrote it looks like zero IF SDR. Perhaps their DSP is not capable of
> >> high sample rates and has a problem with DC offset also.
>
> > No, it's not a zero IF, but it is a low frequency. The last IF is 24
> > kHz. I don't know the sample rate of the ADC.
>
> > Is FM more difficult to do that AM or SSB?
>
> It should work with NBFM, but the sample rate (24 k) is far too
> small for broadcast FM bandwidth.
>
> --
>
> Tauno Voipio, OH2UG

All I want is amateur narrow band - not broadcast.

But I'm puzzled why
 * Kenwood in their TS-590S HF transceiver use a FM demodulator chip
rather than the DSP, wheras they use the DSP for demodulating SSB and
CW.
 * The Flexradio products have a bad reputation for FM demodulation.
I'm guessing they have never cracked how to do it well.

I guess I will just have to try it and find out for myself, though if
anyone has some hard facts comparing different algorithms I'd be
interested.

Dave

David Kirkby wrote:

> On May 5, 7:13 pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> 
>>David Kirkby wrote:
>>
>>>There's one disadvantage for us, in that we intend using a high
>>>quality (24-bit, 192 kbit/s) audio sound-card to do the digitization.
>>
>>Bad idea. That won't work. Stop here.
> 
> I wish you would elaborate more.

Even if the audio card samples at 192kHz, it doesn't mean that it works 
above 20kHz or at low frequencies down to DC. Does this ring a bell?

In addition to that, an audio card creates a good deal of phase 
distortion, which will be translated to nonlinear distortion after FM 
demodulator.

> I think you might be jumping to conclusions.

I think you should start with a textbook, pencil and paper.

> I'm only looking to
> demodulate narrow band FM with a bandwidth of 10 kHz or so. I would
> have thought that perfectly possible with a sample frequency of 192
> KHz. 
> What makes you say an audio card can't be used?

Yes, low IF demodulator could be built. But what is a point?

VLV

>
> > >> I am not familiar with TS-590S architecture details; however from what
> > >> you wrote it looks like zero IF SDR. Perhaps their DSP is not capable of
> > >> high sample rates and has a problem with DC offset also.
>
> > > No, it's not a zero IF, but it is a low frequency. The last IF is 24
> > > kHz. I don't know the sample rate of the ADC.
>
> > > Is FM more difficult to do that AM or SSB?
>
> > It should work with NBFM, but the sample rate (24 k) is far too
> > small for broadcast FM bandwidth.
>
> > --
>
> > Tauno Voipio, OH2UG
>
> All I want is amateur narrow band - not broadcast.
>
> But I'm puzzled why
> �* Kenwood in their TS-590S HF transceiver use a FM demodulator chip
> rather than the DSP, wheras they use the DSP for demodulating SSB and
> CW.
> �* The Flexradio products have a bad reputation for FM demodulation.
> I'm guessing they have never cracked how to do it well.
>
> I guess I will just have to try it and find out for myself, though if
> anyone has some hard facts comparing different algorithms I'd be
> interested.
>
> Dave

if your going to go through the effort to do it in DSP, and it's for
narrow band FM, at least do something interesting like make it a
threshold extension demod.  If you do it in DSP and it works no better
then a conventional limiter discriminator demond then what is the
point?

I've always wanted to see how a good threshold extended demod would
work for narrow band voice FM.

Mark

On 05/05/2011 08:34 AM, David Kirkby wrote:
> I've been looking at making an FM demodulator for a software defined
> radio. In principle at least this is trivial if you have a complex
> signal (I and Q), as you can compute the phase (atan(Q/I), then the
> modulation amplitude is just the derivative of phase with time. As far
> as I can see, this is should trivially easy, though the lack of arctan
> function might be the stumbling block on a fixed point DSP, but thatś
> not an issue for me, as a general purpose x86 CPU will be used.
>
>
> http://www.dsprelated.com/showmessage/50153/1.php
>
> gives a formula:
>
> Vfm = (I*dQ/dt - Q*dI/dt)/(I^2 + Q^2)
>
>
> However, if one looks at many reports of the FlexRadio software
> defined radio, it is clear that FlexRadio have never succeeded in
> getting a decent FM demodulator. The performance of those never seems
> to live up to the hype.
>
> Today I was reading a review in Radio Communication (published by the
> Radio Society of Great Britain) about a Kenwood TS-590S HF transceiver
> covering 1.8 to 30 MHz and also 50 MHz. This is based on a DSP, using
> DSP for demodulation of SSB, variable bandwidth filters, AGC, noise
> reduction etc. But for FM demodulation Kenwood have chosen to convert
> to another IF frequency (455 kHz) and add an FM demodulator chip. All
> they use the DSP for on FM is for audio processing.
>
> Why would a manufacturer of a transceiver go to the bother of adding a
> new IF stage and a FM demodulation chip, when this could be done in
> the DSP they already have?
>
> This is making me think that FM demodulation is a lot more complex
> than I thought it would be.
>
> Has anyone got any comments on the best way to demodulate a mono FM
> signal by using the I and Q outputs of an IQ demodulator?

Hi Dave,

I've watched, with some amusement, all the responses and I don't
think anyone has really hit the mark. Let me take a swing at it.

FM discrimination using the formula you show above, or by using
"standard" differentiation of the phase, involves differentiation.
Differentiation is an inherently noisy process.

One technique which does NOT use differentiation would be a
least-squares estimator of the "instantaneous" frequency. You
do this by finding the phase (arctan) of the I/Q samples, then
fitting the phase values to a line: y = m*x + b. "m" is then
your frequency.

I'm pretty sure this is going to perform better, but you have to be
careful, e.g., you need to carefully evaluate how much of a slice of
time serves for "instantaneous," and to be well-oversampled so that
you get quite a few samples over that slice.
-- 
Randy Yates                      % "So now it's getting late,
Digital Signal Labs              %    and those who hesitate
mailto://yates@ieee.org          %    got no one..."
http://www.digitalsignallabs.com % 'Waterfall', *Face The Music*, ELO

On May 5, 2:34&#4294967295;pm, David Kirkby <drkir...@gmail.com> wrote:

> This is making me think that FM demodulation is a lot more complex
> than I thought it would be.

That's the case with most interesting stuff. That's the reason
why we (still) have colleges and universities to train engineers.
If it was as simple as looking up a web page, anyone could do
engineering. And they can't, do they. Or?

Rune

Mark wrote:


> I've always wanted to see how a good threshold extended demod would
> work for narrow band voice FM.

The ultimate threshold extended demodulator makes the efficient input 
bandwidth equal to the bandwidth of the demodulated signal. Hence the 
gain is input BW/output BW.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

On May 5, 7:44&#4294967295;pm, David Kirkby <drkir...@gmail.com> wrote:
> On May 5, 8:10&#4294967295;pm, Clay <c...@claysturner.com> wrote:
>
> > David, I was working with narrowband FM - i.e., the spectral occupancy
> > was 30kHz. If you are looking at commercial braodcast FM - you are
> > then looking at a much higher spectral occupancy. For example the
> > typical broadcast channel is 200kHz wide so you will need to sample at
> > least 400kHz. The channel is wider than the deviation which in the US
> > is +-75kHz with mono stations being allowed to go an extra 10%. But
> > you can see your samping rate needs to be pretty high.
>
> I'm looking only at narrow band FM for amateur radio use, where
> typically the channel spacing is 12.5 KHz, so onereally needs to keep
> the modulation under that. I'm not interested in commercial FM
> transmissions. Hence I would have thought that using a 24-bit sound
> card able to sample at 192 KHz would have &#4294967295;been suitable.
>
> > You can cheat and narrow the channel down with filtering before
> > demodulating, but then you lose your quieting and then end up with a
> > not so high hi-fi.
>
> > Perhaps the DSP that was already in the receiver was was maxed out
> > with its other tasks. And that is why they added another one.
>
> > IHTH,
> > Clay
>
> Perhaps that's the reason, but the fact I know the FlexRadio has FM
> demodulation issues makes me wonder if demodulating even narrow band
> FM is more difficult than I thought it would be.
>
> Dave

Hello Dave,

When I demodulated narrowband FM, I did (as Vladimir suggests) and
made my analytic signal in software. Yes this does burn some cycles,
but then you can achieve nearly perfect 90 degree phase difference and
matching amplitudes across the band of interest. THe IdQ-QDi approach
can actually work pretty well if you do the right approximations.

Clay

On May 6, 12:51&#4294967295;pm, Clay <c...@claysturner.com> wrote:
> On May 5, 7:44&#4294967295;pm, David Kirkby <drkir...@gmail.com> wrote:
>
>
>
>
>
> > On May 5, 8:10&#4294967295;pm, Clay <c...@claysturner.com> wrote:
>
> > > David, I was working with narrowband FM - i.e., the spectral occupancy
> > > was 30kHz. If you are looking at commercial braodcast FM - you are
> > > then looking at a much higher spectral occupancy. For example the
> > > typical broadcast channel is 200kHz wide so you will need to sample at
> > > least 400kHz. The channel is wider than the deviation which in the US
> > > is +-75kHz with mono stations being allowed to go an extra 10%. But
> > > you can see your samping rate needs to be pretty high.
>
> > I'm looking only at narrow band FM for amateur radio use, where
> > typically the channel spacing is 12.5 KHz, so onereally needs to keep
> > the modulation under that. I'm not interested in commercial FM
> > transmissions. Hence I would have thought that using a 24-bit sound
> > card able to sample at 192 KHz would have &#4294967295;been suitable.
>
> > > You can cheat and narrow the channel down with filtering before
> > > demodulating, but then you lose your quieting and then end up with a
> > > not so high hi-fi.
>
> > > Perhaps the DSP that was already in the receiver was was maxed out
> > > with its other tasks. And that is why they added another one.
>
> > > IHTH,
> > > Clay
>
> > Perhaps that's the reason, but the fact I know the FlexRadio has FM
> > demodulation issues makes me wonder if demodulating even narrow band
> > FM is more difficult than I thought it would be.
>
> > Dave
>
> Hello Dave,
>
> When I demodulated narrowband FM, I did (as Vladimir suggests) and
> made my analytic signal in software. Yes this does burn some cycles,
> but then you can achieve nearly perfect 90 degree phase difference and
> matching amplitudes across the band of interest. THe IdQ-QDi approach
> can actually work pretty well if you do the right approximations.
>
> Clay- Hide quoted text -
>
> - Show quoted text -

Dave,

See

http://www.claysturner.com/dsp/instantaneousfrequency.pdf

for one example of a simplified form of the cross correlation
approach. It uses low order differentiators so it won't try to magnify
the noise too much.

IHTH,
Clay

On May 6, 11:00&#4294967295;am, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> Mark wrote:
> > I've always wanted to see how a good threshold extended demod would
> > work for narrow band voice FM.
>
> The ultimate threshold extended demodulator makes the efficient input
> bandwidth equal to the bandwidth of the demodulated signal. Hence the
> gain is input BW/output BW.
>
> Vladimir Vassilevsky
> DSP and Mixed Signal Design Consultanthttp://www.abvolt.com

conventional "communications narrow band voice FM"  uses +/- 5 kHz
deviation with an RF BW of about 13 kHz and a demodulated BW of about
3 kHz so some worth while "extension" might be possible.

It would be interesting to see how it works in practice.

I thought it would be interesting to combine a very narrow band
threshold extended demod that was as narrow as possible to optimize
recovered SNR at the expense of some distortion with a conventional
demod that would automatically switch in whenever  the signal was
above threshold.

Of course the world has passed analog FM by...

Mark

Mark wrote:

> On May 6, 11:00 am, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> 
>>Mark wrote:
>>
>>>I've always wanted to see how a good threshold extended demod would
>>>work for narrow band voice FM.
>>
>>The ultimate threshold extended demodulator makes the efficient input
>>bandwidth equal to the bandwidth of the demodulated signal. Hence the
>>gain is input BW/output BW.
>>
> conventional "communications narrow band voice FM"  uses +/- 5 kHz
> deviation with an RF BW of about 13 kHz

Actually there are two widely accepted variants of narrow FM: used with 
the channel spacing of 12.5kHz and 25kHz correspondingly. The modulation 
parameters and the RF bandwidth are somewhat different.

> and a demodulated BW of about
> 3 kHz so some worth while "extension" might be possible.
> 
> It would be interesting to see how it works in practice.
> 
> I thought it would be interesting to combine a very narrow band
> threshold extended demod that was as narrow as possible to optimize
> recovered SNR at the expense of some distortion with a conventional
> demod that would automatically switch in whenever  the signal was
> above threshold.

I thought many times about the optimal FM demodulator for the voiceband 
communications. It would be an interesting DSP project, and it could 
have practical application. I would be happy to develop such product if 
someone sponsors my effort.

> Of course the world has passed analog FM by...

Not really; there are still zillions of analog FM channels arround.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com