How do you detect the broadband noise level?

The traditional diode detector does not distinguish between broadband
noise and narrowband signals.

Is there a way to detect the overall broadband noise level?  In this
coming semester, I will be working on an automatic noise cancelling
device for amateur radio.  The input for the feedback network will be
provided by a circuit that responds to broadband noise but does not
respond to narrowband signals.  I would prefer to monitor the noise at
the RF level (BEFORE the signals and noise reach the receiver), but it
would probably be easier and cheaper to monitor the audio noise.

The only idea I can think of is using an A/D converter to convert the
audio output of the receiver into discrete-time signals, using DSP to
convert these signals into a DFT, and then integrating over the DFT to
come up with the noise floor measurement.  Of course, there needs to
be a way to take the absolute value of the discrete points.

Has anyone here done anything like this?  I studied DSP last semester,
but we never touched a single DSP chip in the class.  Using and
implementing DSP hardware is something I'll be picking up on my own.

Jason Hsu, AG4DG
usenet@@@@jasonhsu.com

"Jason Hsu" <jason_hsu@my-deja.com> wrote in message
news:f7d9a152.0308201652.7837c112@posting.google.com...
> The traditional diode detector does not distinguish between broadband
> noise and narrowband signals.
>
> Is there a way to detect the overall broadband noise level?  In this
> coming semester, I will be working on an automatic noise cancelling
> device for amateur radio.  The input for the feedback network will be
> provided by a circuit that responds to broadband noise but does not
> respond to narrowband signals.  I would prefer to monitor the noise at
> the RF level (BEFORE the signals and noise reach the receiver), but it
> would probably be easier and cheaper to monitor the audio noise.
>
> The only idea I can think of is using an A/D converter to convert the
> audio output of the receiver into discrete-time signals, using DSP to
> convert these signals into a DFT, and then integrating over the DFT to
> come up with the noise floor measurement.  Of course, there needs to
> be a way to take the absolute value of the discrete points.
>
> Has anyone here done anything like this?  I studied DSP last semester,
> but we never touched a single DSP chip in the class.  Using and
> implementing DSP hardware is something I'll be picking up on my own.

Jason,

You haven't really told us what the block diagram is - so providing advice
is a bit difficult.
Perhaps you're thinking of something like a line enhancer?

In many cases, the broadband noise is dominant if there are only a few
narrowband signals - until the SNR gets to be rather large.  Is the
narrowband signal always present or only intermittently?  Is the broadband
noise of constant amplitude/energy or very slowly changing - as is often the
case?

Once you know the noise level, what are you going to do with it?  Or, is
this a squelch circuit that simply cuts off when there's no signal?

Fred

jason_hsu@my-deja.com (Jason Hsu) wrote in message news:<f7d9a152.0308201652.7837c112@posting.google.com>...
> The traditional diode detector does not distinguish between broadband
> noise and narrowband signals.
> 
> Is there a way to detect the overall broadband noise level?  In this
> coming semester, I will be working on an automatic noise cancelling
> device for amateur radio.  The input for the feedback network will be
> provided by a circuit that responds to broadband noise but does not
> respond to narrowband signals.  I would prefer to monitor the noise at
> the RF level (BEFORE the signals and noise reach the receiver), but it
> would probably be easier and cheaper to monitor the audio noise.
> 
> The only idea I can think of is using an A/D converter to convert the
> audio output of the receiver into discrete-time signals, using DSP to
> convert these signals into a DFT, and then integrating over the DFT to
> come up with the noise floor measurement.  Of course, there needs to
> be a way to take the absolute value of the discrete points.
> 
> Has anyone here done anything like this?  I studied DSP last semester,
> but we never touched a single DSP chip in the class.  Using and
> implementing DSP hardware is something I'll be picking up on my own.

If you really do want to measure the RF signal level at the input of
your receiver, it might be worth your while looking at the Analog
Devices AD361 or AD362 broad-band true rms detectors.

------
Bill Sloman, Nijmegen

Jason Hsu wrote:

> The traditional diode detector does not distinguish between broadband
> noise and narrowband signals.
>
> Is there a way to detect the overall broadband noise level?  In this
> coming semester, I will be working on an automatic noise cancelling
> device for amateur radio.  The input for the feedback network will be
> provided by a circuit that responds to broadband noise but does not
> respond to narrowband signals.  I would prefer to monitor the noise at
> the RF level (BEFORE the signals and noise reach the receiver), but it
> would probably be easier and cheaper to monitor the audio noise.
>
> The only idea I can think of is using an A/D converter to convert the
> audio output of the receiver into discrete-time signals, using DSP to
> convert these signals into a DFT, and then integrating over the DFT to
> come up with the noise floor measurement.  Of course, there needs to
> be a way to take the absolute value of the discrete points.
>
> Has anyone here done anything like this?  I studied DSP last semester,
> but we never touched a single DSP chip in the class.  Using and
> implementing DSP hardware is something I'll be picking up on my own.
>
> Jason Hsu, AG4DG
> usenet@@@@jasonhsu.com

I think this requires a DSP solution. Do a analog to digital conversion
and then analyse the frequency content and reject those components
that you think are narrow band signals. Take the rest as "noise"

Good luck.
Bill K7NOM

Jason,
Look at your signal level.  Determine when the signal level is
smallest.  Remember that if your received signal is desired + noise,
when the desired is absent, you have noise.  This "smallest" signal
level measurement will give an approximation of the noise.

Maurice Givens


jason_hsu@my-deja.com (Jason Hsu) wrote in message news:<f7d9a152.0308201652.7837c112@posting.google.com>...
> The traditional diode detector does not distinguish between broadband
> noise and narrowband signals.
> 
> Is there a way to detect the overall broadband noise level?  In this
> coming semester, I will be working on an automatic noise cancelling
> device for amateur radio.  The input for the feedback network will be
> provided by a circuit that responds to broadband noise but does not
> respond to narrowband signals.  I would prefer to monitor the noise at
> the RF level (BEFORE the signals and noise reach the receiver), but it
> would probably be easier and cheaper to monitor the audio noise.
> 
> The only idea I can think of is using an A/D converter to convert the
> audio output of the receiver into discrete-time signals, using DSP to
> convert these signals into a DFT, and then integrating over the DFT to
> come up with the noise floor measurement.  Of course, there needs to
> be a way to take the absolute value of the discrete points.
> 
> Has anyone here done anything like this?  I studied DSP last semester,
> but we never touched a single DSP chip in the class.  Using and
> implementing DSP hardware is something I'll be picking up on my own.
> 
> Jason Hsu, AG4DG
> usenet@@@@jasonhsu.com

jason_hsu@my-deja.com (Jason Hsu) wrote in message news:<f7d9a152.0308201652.7837c112@posting.google.com>...
> The traditional diode detector does not distinguish between broadband
> noise and narrowband signals.
> 
> Is there a way to detect the overall broadband noise level?  In this
> coming semester, I will be working on an automatic noise cancelling
> device for amateur radio.  The input for the feedback network will be
> provided by a circuit that responds to broadband noise but does not
> respond to narrowband signals.  I would prefer to monitor the noise at
> the RF level (BEFORE the signals and noise reach the receiver), but it
> would probably be easier and cheaper to monitor the audio noise.
> 
> The only idea I can think of is using an A/D converter to convert the
> audio output of the receiver into discrete-time signals, using DSP to
> convert these signals into a DFT, and then integrating over the DFT to
> come up with the noise floor measurement.  Of course, there needs to
> be a way to take the absolute value of the discrete points.
> 
> Has anyone here done anything like this?  I studied DSP last semester,
> but we never touched a single DSP chip in the class.  Using and
> implementing DSP hardware is something I'll be picking up on my own.
> 
> Jason Hsu, AG4DG
> usenet@@@@jasonhsu.com


First you realize that you can only truly "cancel" noise that is
correlated - white noise and most "natural" noise source are random
and uncorrelated to anything including themselves (no
autocorrelation).  Most devices sold colloquially as "noise
cancellers" can only work on periodic noise sources (like prop or fan
noise in planes), noise signals which have an autocorrelation.

For most RF/uW signal the only controllable variables for noise are BW
and T (Pn = kT * BW) which is why RF and IF filter bandwidth figures
so highly in RF receiver design.  The noise contribution of the
atmosphere/ionosphere/etc. (aka channel noise) is random.  Trying to
substract noise actually adds noise into the signal because you must
use precisely 180 degree out-of-phase, equal magnitude canceling
signal derived from a measured signal but being uncorrelated means
that both phase and magnitude are unpredictable from previous
measurements thus you can't be sure you are adding 0 or 180 degree
cancellation.  After that your only choice is better signal (better
antennas) to punch through the limits of what ever S/N or noise figure
your cable and receiver can give you. This is why the ARRL has long
had the nice and fat Antenna Handbook available.

I think true "canceling" noise will be problematic but you could use
knowing the noise floor to adaptively change the receiver bandwidth to
"optimize net S/N" at the price of some signal fidelity.  For example
set BW = k * (S - Nref), i.e. open up the IF when the signal is well
above the noise floor and tighten it up when the gap shrinks. You can
measure/characterize your system's noise characteristics and then
infer the input noise from that if there is no signal at the input. 
The latter is like your idea of integrating out-of-band levels.  The
implicit assumption is that there is no signal out-of-band which might
be tricky in some Ham bands.  BTW noise canceller functions on RF
receivers are often doing something like this.

Alternatively since most (except for highly compressed or encrypted
signals) have autocorrelations if you have a reference for the signal
you can reinforce autocorrelated signal and diminish
non-autocorrelated signal aka noise.  This works best with digital
modulation.  To get a usable reference you need a clean version of
your signal available - obvious a problem since that is what the
receiver is trying to do in the first place.  The trick is diversity:
find a copy of the signal that took a different path spatially, in
time or in frequency whose degradation is not cross-correlated to the
first signal.  You can use multiple antennas in different locations
(many wavelength-wise distant) - ArrayComm is doing this for cellular
telephones using phased array antennas.  Spread spectrum creates
diversity mostly over frequency and somewhat over time where the
demodulation is implicitly "removing" uncorrelated noise.  This whole
discussion is a large part of why we are really moving to digital
modulation standards - you need a "clean" reference for the actually
signal to remove channel noise, and there's only some much you can do
to recover a "clean" reference to cancel noise if the signal is
analog.  ("clean" meaning clean enough to remove uncorrelated noise).

Throwing an ADC on the whole thing complicates things - the
quantization noise added by an ADC itself could easily swamp the noise
signal you are trying to measure if you don't design the analog
section properly.  Just be careful.

They probably never covered any of this truly juicy stuff in your DSP
class - academics tend to not be close enough to reality to get out of
just pushing the equations.

MM

Mantra wrote:
> 
> jason_hsu@my-deja.com (Jason Hsu) wrote in message news:<f7d9a152.0308201652.7837c112@posting.google.com>...
> > The traditional diode detector does not distinguish between broadband
> > noise and narrowband signals.
> >
> > Is there a way to detect the overall broadband noise level?  In this
> > coming semester, I will be working on an automatic noise cancelling
> > device for amateur radio.  The input for the feedback network will be
> > provided by a circuit that responds to broadband noise but does not
> > respond to narrowband signals.  I would prefer to monitor the noise at
> > the RF level (BEFORE the signals and noise reach the receiver), but it
> > would probably be easier and cheaper to monitor the audio noise.
> >
> > The only idea I can think of is using an A/D converter to convert the
> > audio output of the receiver into discrete-time signals, using DSP to
> > convert these signals into a DFT, and then integrating over the DFT to
> > come up with the noise floor measurement.  Of course, there needs to
> > be a way to take the absolute value of the discrete points.
> 
> First you realize that you can only truly "cancel" noise that is
> correlated - white noise and most "natural" noise source are random
> and uncorrelated to anything including themselves (no
> autocorrelation).  Most devices sold colloquially as "noise
> cancellers" can only work on periodic noise sources (like prop or fan
> noise in planes), noise signals which have an autocorrelation.
> 

<reminiscence>This has been being done very successfully since about 1965 that I know
about. It takes one tube if you do it the 1965 way.</reminiscence>

It's even simpler than that--most of the broadband noise you want to reject in a
receiver is impulse noise, which consists of a train of large spikes.  What you do is
to put a Schmitt trigger before the first IF filter, looking for large pulses, and
use a retriggerable monostable to turn off the IF gain until a few settling times
after the pulse has gone away again.  The delay in the IF filter guarantees that the
gain will be off by the time the pulse arrives at the IF amplifier.  Shorting out the
input to the IF filter is even better, because you don't have to wait for the filter
to settle afterwards.

There's no point in trying to do this at AF or even following the IF filter, because
a narrowband filter turns a big narrow spike into a long tone burst, at which point
it's a good deal harder to get rid of.

Using a DSP correlator at the audio output is more or less like trying to undo the
action of the IF filter, which seems a bit pointless when you can get the same effect
by moving the wire from one place to another.  Another way of saying this is that the
autocorrelation of wideband noise becomes quite similar to the autocorrelation of the
signal, once you pass the IF filter, so there's less you can do about it.

Cheers,

Phil Hobbs