SNR estimation before or after matched filtering

I'll try to summarise what I understood. 
There are two aspects covered in the above discussion. 
1. A way to determine the SNR threshold, i.e., minimum SNR for system to work. There are two suggested ways:
a. To transmit in an ideal environment. And then generate noise with known power, increase noise till the receiver givesup. 
b. To transmit in a noisy environment. Measure noise and estimate noise power at the Rx. Start decreasing the Tx power till the receiver is unable to hear the drowned signal. 

2. How to define SNR? 
This is tricky, cus only the inband noise power must be considered when measuring the SNR value, i.e., considering 1a I'll take the power of the generated noise after the matched filter, i.e, sigma_n |H(w)|^2 and then compute S/N.  

Fred Marshall wrote:

> On 10/30/2011 8:57 AM, Eric Jacobsen wrote:
> 
>> On Sun, 30 Oct 2011 03:53:04 -0700 (PDT), mobi<mobien@gmail.com>
>> wrote:
>>
>>> I need to determine minimum working SNR for transceiver system over a 
>>> fixed distance. I change the Tx power till the Rx cannot make sense 
>>> of the message. The problem is to estimate the SNR at the receiver. 

> Assume that the receiver prefiltered output noise is broadband Gaussian, 
> therefore white in the passband of the receiver.  Often this is a good 
> model anyway.
> And, if this assumption isn't a good one, you can figure that out later.

Bla-bla-bla.

> The result should be readily estimated as well as measured.

Nope. There is a bazilion of factors difficult to account for, from the 
receiver nonlinearity to the channel spread. The way to measure the S/N 
and sensitivity correctly is by the use of a noise generator. This is a 
standard practice.


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

 Lemme try to summarize what i understood. 

Basically there are two aspects covered in the above discussion. 

1. Issue of determining the SNR threhsold. 

Two suggested solutions are: 

a. Setup the Tx and Rx in ideal environment. Tx power in known. Generate noise using noise generator. Increase the noise power to drown the Tx signal to effect that Rx cannot rescue. Noise power is known. 

b. Tx and Rx is setup. Measure the noise alone (No Tx signal). Start decreasing Tx signal till Rx cannot resolve it. But I've to make sure that I've the dynamic range of Rx in picture while following this method. 

2. Issue of determining the SNR. 

Inband noise power must be considered only. So for 1a. I need to adjust the noise power w.r.t the matched filter and then compute S/N. 

 Lemme try to summarize what i understood. 

Basically there are two aspects covered in the above discussion. 

1. Issue of determining the SNR threhsold. 

Two suggested solutions are: 

a. Setup the Tx and Rx in ideal environment. Tx power in known. Generate noise using noise generator. Increase the noise power to drown the Tx signal to effect that Rx cannot rescue. Noise power is known. 

b. Tx and Rx is setup. Measure the noise alone (No Tx signal). Start decreasing Tx signal till Rx cannot resolve it. But I've to make sure that I've the dynamic range of Rx in picture while following this method. 

2. Issue of determining the SNR. 

Inband noise power must be considered only. So for 1a. I need to adjust the noise power w.r.t the matched filter and then compute S/N. 

> Lemme try to summarize what i understood. 
>
>Basically there are two aspects covered in the above discussion. 
>
>1. Issue of determining the SNR threhsold. 
>
>Two suggested solutions are: 
>
>a. Setup the Tx and Rx in ideal environment. Tx power in known. Generate
noise using noise generator. Increase the noise power to drown the Tx
signal to effect that Rx cannot rescue. Noise power is known. 
>
>b. Tx and Rx is setup. Measure the noise alone (No Tx signal). Start
decreasing Tx signal till Rx cannot resolve it. But I've to make sure that
I've the dynamic range of Rx in picture while following this method. 
>
>2. Issue of determining the SNR. 
>
>Inband noise power must be considered only. So for 1a. I need to adjust
the noise power w.r.t the matched filter and then compute S/N. 
>

For threshold this is what I wil do:
add known noise say digitaly at Tx and keep lowering SNR checking Rx BER.
This is straightforward.

If the issue is SNR measurement at received signal i.e. SNR of channel then
capture your Rx input at start and do frequency analysis to measure SNR but
for this measurement a single tone is better.

I don't see why you have to look at Rx filtered outputs if you want to
estimate channel noise.

Kadhiem

On Sun, 30 Oct 2011 14:45:15 -0700 (PDT), mobi <mobien@gmail.com>
wrote:

>I'll try to summarise what I understood. 
>There are two aspects covered in the above discussion. 
>1. A way to determine the SNR threshold, i.e., minimum SNR for system to work. There are two suggested ways:
>a. To transmit in an ideal environment. And then generate noise with known power, increase noise till the receiver givesup. 

This is for bench testing of things like FEC performance,
implementation loss, etc.   It will require some means to calibrate
both the noise and power measurements, etc., etc., probably including
a means to measure either BER or PER or some other useful objective
performance reliability metric.

You don't need any on-board, i.e., in-receiver measurements or
computations to do these tests other than a means to measure
BER/PER/whatever.

>b. To transmit in a noisy environment. Measure noise and estimate noise power at the Rx. Start decreasing the Tx power till the receiver is unable to hear the drowned signal. 

This is for system threshold testing or for reporting estimated
performance in the field or during deployment.  In addition to what
was measured in 1a will also test the receiver floor characteristics.

They're different tests, they're both useful, they're both commonly
done, the test slightly different things.

>2. How to define SNR? 
>This is tricky, cus only the inband noise power must be considered when measuring the SNR value, i.e., considering 1a I'll take the power of the generated noise after the matched filter, i.e, sigma_n |H(w)|^2 and then compute S/N.  

Defining SNR can be tricky and may be different for different systems
especially depending on whether it is packet based or continuous or
has multipath or not, etc., etc.    For a continuous AWGN system it's
not that bad but may still depend on how the system specs were
determined, etc., etc.


Eric Jacobsen
Anchor Hill Communications
www.anchorhill.com

>I need to determine minimum working SNR for transceiver system over a
fixed distance. I change the Tx power till the Rx cannot make sense of the
message. The problem is to estimate the SNR at the receiver. I've two main
questions
>1. Is it possible to estimate the SNR reliably from the spectrum of the
received signal? 
>2. Should the SNR be computed before matched filtering or after? 
>

1): Yes:
- measure power in the absence of signal (noise only)
- measure power in the presence of signal (signal plus noise)
- subtract => signal power only
- divide signal by noise => SNR

How to measure power, see 2): we do need some kind of filter...

2) Yes, a matched filter is a good choice, if you're free to chose. That
is, if no other definition is given.

The resulting SNR would be applicable to an optimal linear receiver. BUT,
depending on the application, technology may have moved already beyond
"optimal linear receiver" (think interference cancellation on adjacent
channel). I'm just mentioning this only to point out that there is no
single correct answer, many devils are hiding in the details. A matched
filter is probably a good choice, though.