Tin can modulation, OFDM, PAPR

As was kindly suggested to me, I've been experimenting with OFDM for my
TCP/IP over tin cans project. Since the modulation on the "wire" is at
baseband, and I'm using a sound card as a DAC, this means my signals
are audible.

When modulating a random bitstream, the resulting audio is full of
popping noises. I have taken steps to ensure that no clipping is
occurring, so these pops are really a part of the signal, not nonlinear
effects. Am I hearing the effects of OFDM's high PAPR?

It's true that this channel is very dispersive, but the characteristics
should be constant with time, because I am trying to idealize the
implementation. As long as the channel is reasonably linear, shouldn't
I be able to counter this dispersion by deconvolution? I realize that
the FIR kernel might be quite large. If that's possible, I'd prefer to
use some other modulation with better PAPR characteristics.

Also, any ideas how I might measure the degree of linearity of the
system? I was thinking of sweeping a band-pass filter across some white
noise and looking at the spectrum at the receiver. Any nonlinearity
should manifest as energy outside of the original band limits, correct?
I suppose the nonlinearity could somehow stay entirely within the band
(imagine randomly reordering the values of the FFT bins, but staying
within the band limits) but I think the chances of that being the case
are fairly remote.

Thanks,
Scott

Hi Scott,

Seems Don Quixote is alive and well. :-)

Scott L wrote:
> As was kindly suggested to me, I've been experimenting with OFDM for my
> TCP/IP over tin cans project. Since the modulation on the "wire" is at
> baseband, and I'm using a sound card as a DAC, this means my signals
> are audible.
> 
> When modulating a random bitstream, the resulting audio is full of
> popping noises. I have taken steps to ensure that no clipping is
> occurring, so these pops are really a part of the signal, not nonlinear
> effects. Am I hearing the effects of OFDM's high PAPR?
> 
> It's true that this channel is very dispersive, but the characteristics

A good way to reduce dispersion might be to choose the string wisely. A 
monofilament should improve things. Say, a thin fibre of glass, held 
under light tension. You don't usually see people trying OFDM over fibre :-)

> should be constant with time, because I am trying to idealize the
> implementation. As long as the channel is reasonably linear, shouldn't
> I be able to counter this dispersion by deconvolution? I realize that
> the FIR kernel might be quite large. If that's possible, I'd prefer to
> use some other modulation with better PAPR characteristics.
> 
> Also, any ideas how I might measure the degree of linearity of the
> system? I was thinking of sweeping a band-pass filter across some white
> noise and looking at the spectrum at the receiver. Any nonlinearity
> should manifest as energy outside of the original band limits, correct?
> I suppose the nonlinearity could somehow stay entirely within the band
> (imagine randomly reordering the values of the FFT bins, but staying
> within the band limits) but I think the chances of that being the case
> are fairly remote.

Regards,
Steve

Scott L wrote:
> As was kindly suggested to me, I've been experimenting with OFDM for my
> TCP/IP over tin cans project. Since the modulation on the "wire" is at
> baseband, and I'm using a sound card as a DAC, this means my signals
> are audible.
> 
> When modulating a random bitstream, the resulting audio is full of
> popping noises. I have taken steps to ensure that no clipping is
> occurring, so these pops are really a part of the signal, not nonlinear
> effects. Am I hearing the effects of OFDM's high PAPR?
> 
> It's true that this channel is very dispersive, but the characteristics
> should be constant with time, because I am trying to idealize the
> implementation. As long as the channel is reasonably linear, shouldn't
> I be able to counter this dispersion by deconvolution? I realize that
> the FIR kernel might be quite large. If that's possible, I'd prefer to
> use some other modulation with better PAPR characteristics.
> 
> Also, any ideas how I might measure the degree of linearity of the
> system? I was thinking of sweeping a band-pass filter across some white
> noise and looking at the spectrum at the receiver. Any nonlinearity
> should manifest as energy outside of the original band limits, correct?
> I suppose the nonlinearity could somehow stay entirely within the band
> (imagine randomly reordering the values of the FFT bins, but staying
> within the band limits) but I think the chances of that being the case
> are fairly remote.
> 
> Thanks,
> Scott
> 
AFAIK the 'pop' can be quite smeared, as long as it doesn't get 
distorted by a nonlinearity (which would spill energy into other 
symbols), and as long as the dispersion is a smallish fraction of the 
symbol time of your OFDM.

If you're actually doing this, why don't you try measuring the 
dispersion of the line?  Give it a nice narrow but bandlimited pulse, 
like a time-limited sync pulse or a Gaussian or some such, and see what 
comes out the other end.

-- 

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

Posting from Google?  See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

On Wed, 24 Jan 2007 20:20:24 +0800, Steve Underwood <steveu@dis.org>
wrote:

>Hi Scott,
>
>Seems Don Quixote is alive and well. :-)
>
>Scott L wrote:
>> As was kindly suggested to me, I've been experimenting with OFDM for my
>> TCP/IP over tin cans project. Since the modulation on the "wire" is at
>> baseband, and I'm using a sound card as a DAC, this means my signals
>> are audible.
>> 
>> When modulating a random bitstream, the resulting audio is full of
>> popping noises. I have taken steps to ensure that no clipping is
>> occurring, so these pops are really a part of the signal, not nonlinear
>> effects. Am I hearing the effects of OFDM's high PAPR?
>> 
>> It's true that this channel is very dispersive, but the characteristics
>
>A good way to reduce dispersion might be to choose the string wisely. A 
>monofilament should improve things. Say, a thin fibre of glass, held 
>under light tension. You don't usually see people trying OFDM over fibre :-)

Actually there's a move to multiple-wavelength transmission on fibers,
but I think you're right that they're not orthogonal.
Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org

On 23 Jan 2007 16:56:20 -0800, "Scott L" <scott-sp02@neuralnw.com>
wrote:

>As was kindly suggested to me, I've been experimenting with OFDM for my
>TCP/IP over tin cans project. Since the modulation on the "wire" is at
>baseband, and I'm using a sound card as a DAC, this means my signals
>are audible.
>
>When modulating a random bitstream, the resulting audio is full of
>popping noises. I have taken steps to ensure that no clipping is
>occurring, so these pops are really a part of the signal, not nonlinear
>effects. Am I hearing the effects of OFDM's high PAPR?

Possibly, and if so, are you using a scrambler?   Just whitening the
data goes a long ways toward reducing PAPR in OFDM.   Also look
carefully at the arrangement of any pilot tones or preamble/channel
estimation sequences that you are using, as they need to be considered
for PAPR as well.

>It's true that this channel is very dispersive, but the characteristics
>should be constant with time, because I am trying to idealize the
>implementation. As long as the channel is reasonably linear, shouldn't
>I be able to counter this dispersion by deconvolution? I realize that
>the FIR kernel might be quite large. If that's possible, I'd prefer to
>use some other modulation with better PAPR characteristics.

If the OFDM symbol rate is low enough it may be that a cyclic-prefix
is all that is needed.  I guess I'm not completely seeing the
connection between the dispersion and PAPR?

In any case, I agree with the previous posts that characterizing the
channel transfer function is a good idea.   I think I'd look at it
with different string types, string tensions, can diameters, etc. You
could spend a ton of time just on channel characterization, though,
and you probably want to get on with the system design.

>Also, any ideas how I might measure the degree of linearity of the
>system? I was thinking of sweeping a band-pass filter across some white
>noise and looking at the spectrum at the receiver. Any nonlinearity
>should manifest as energy outside of the original band limits, correct?
>I suppose the nonlinearity could somehow stay entirely within the band
>(imagine randomly reordering the values of the FFT bins, but staying
>within the band limits) but I think the chances of that being the case
>are fairly remote.

Rather than a BPF sweep with noise, why not sweep a linear FM and
recover magnitude *and* phase in the receiver.   That gives about as
complete of a picture as possible.   For nonlinearity assessment you
may want to do that at increasing power levels, i.e., FM sweep,
increase power, FM sweep, increase power, sweep again, etc.  At some
point any nonlinearity and it's characteristic will become apparent.

Or, to save time, you could do a single FM sweep and pick a few
frequencies to test based on the results.   Start a tone at a
frequency of interest and gradually increase the power and see what
happens.

I think it's an interesting problem and just getting good
characterizations of the channel and the effects of the string types,
tension, and can types, diameters, etc.,  would be interesting.

Not real useful, but interesting anyway...   ;)

Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org

On Jan 24, 11:13 am, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
> >When modulating a random bitstream, the resulting audio is full of
> >popping noises. I have taken steps to ensure that no clipping is
> >occurring, so these pops are really a part of the signal, not nonlinear
> >effects. Am I hearing the effects of OFDM's high PAPR?

> Possibly, and if so, are you using a scrambler?   Just whitening the
> data goes a long ways toward reducing PAPR in OFDM.   Also look
> carefully at the arrangement of any pilot tones or preamble/channel
> estimation sequences that you are using, as they need to be considered
> for PAPR as well.

There is no scrambler or coding at this stage. My first attempt was a
modulation of several kb of straight ASCII text, which has a fairly low
entropy (and a few bit magnitudes which never change their values!).
When I moved to random data as the input, the popping was greatly
reduced but still apparent.

With no mathematical justification whatsoever, I tried using different
base phases on each subcarrier (right now the carrier modulation is
BPSK). This made a difference but not much.

> >It's true that this channel is very dispersive, but the characteristics
> >should be constant with time, because I am trying to idealize the
> >implementation. As long as the channel is reasonably linear, shouldn't
> >I be able to counter this dispersion by deconvolution? I realize that
> >the FIR kernel might be quite large. If that's possible, I'd prefer to
> >use some other modulation with better PAPR characteristics.

> If the OFDM symbol rate is low enough it may be that a cyclic-prefix
> is all that is needed.  I guess I'm not completely seeing the
> connection between the dispersion and PAPR?

It's not really a connection. What I meant is that OFDM was suggested
because of its good performance on dispersive channels. If I can remove
the dispersion by deconvolution then I don't have to deal with it and
that gives me more options besides OFDM. Or is this wrong thinking?

As for a cyclic prefix... I can't experiment usefully with that until I
know the appropriate guard interval to use, which I can't determine yet
because my rig isn't built.

> In any case, I agree with the previous posts that characterizing the
> channel transfer function is a good idea.   I think I'd look at it
> with different string types, string tensions, can diameters, etc. You
> could spend a ton of time just on channel characterization, though,
> and you probably want to get on with the system design.

Yes, I want to do this as soon as possible. My biggest problem at the
moment is shielding the transmitter and receiver so that the receiver
cannot hear the sound directly through the air -- I want it only to
hear the signal coming from the can. I'm lacking the materials to do
this properly so I have to improvise.

I'm having trouble finding a small speaker that can fit right inside
the mouth of the can and also matches the impedance of the sound card
audio output. And the mic I'm working with currently is cheap.

> >Also, any ideas how I might measure the degree of linearity of the
> >system? I was thinking of sweeping a band-pass filter across some white
> >noise and looking at the spectrum at the receiver. Any nonlinearity
> >should manifest as energy outside of the original band limits, correct?
> >I suppose the nonlinearity could somehow stay entirely within the band
> >(imagine randomly reordering the values of the FFT bins, but staying
> >within the band limits) but I think the chances of that being the case
> >are fairly remote.

> Rather than a BPF sweep with noise, why not sweep a linear FM and
> recover magnitude *and* phase in the receiver.   That gives about as
> complete of a picture as possible.   For nonlinearity assessment you
> may want to do that at increasing power levels, i.e., FM sweep,
> increase power, FM sweep, increase power, sweep again, etc.  At some
> point any nonlinearity and it's characteristic will become apparent.
>
> Or, to save time, you could do a single FM sweep and pick a few
> frequencies to test based on the results.   Start a tone at a
> frequency of interest and gradually increase the power and see what
> happens.

Good suggestions, thanks.

> Not real useful, but interesting anyway...   ;)

Steve Underwood's characterization of me as "Don Quixote" is fairly
accurate...

Thanks again,
Scott