Message-ID: <3AC1AFA8.894F6591@usa.net>
Date: Wed, 28 Mar 2001 11:32:24 +0200
From: "Alberto di Bene" <dibene@usa.net>
Organization: Undisclosed
MIME-Version: 1.0
To: rsgb_lf_group@blacksheep.org
Subject: LF: Re: [TECH] A question
References: <65AECDF1F89AD411900400508BFC869F9C0407@pdw-mail-1.dera.gov.uk>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk
Reply-To: rsgb_lf_group@blacksheep.org
Sender: <majordom@post.thorcom.com>

Andy,
          thanks for your suggestion. As a matter of fact I am already using
your algorithm ! I use it to compute the brightness levels for the spectrogram,
but only when AGC is on. So, with AGC on, the brightness is a (non-linear)
function of the ratio between the magnitude and the base noise level
computed with the CFAR method. I learnt it from your previous post of many
months ago, I liked it, and I put it into immediate use.
I will use it when implementing the S/N display, together with the choice of
selecting first a frequency range, whose mean level will be the noise reference
level for the computation.
For the present problem, displaying a relative strenght level, I think I will
allow the choice between the two main methods that have been suggested,
i.e. referring it either to the saturation value of the ADC, or to a previously
choosen reference level.

73  Alberto  I2PHD

Talbot Andrew wrote:

> One method for determining S/N ration in an environment where signals
> and noise are constantly changing  - used a lot in Radar and Other
> techniques - is an algorithm called Constant False Alarm Rate.   I
> covered this in detail in postings to this reflector several months ago,
> but basically the technique is as follows :
>
> 1)      Perform an FFT on a block of data
> 2)      Sort all the bins into increasing order of amplitude.
> 3)      Take the amplitude of the lower quartile bin.
> 4)      Add 3dB to this figure and this gives a very good approximation
> to the noise level
> 5)      For new signal alarm and signal detection add a threshold - 10dB
> is a good starting point
> 6)      Check for successive hits above this threshold N out of M times
> before indicating valid.
>
> You will only be interested in satges 1) to 4).  This assumes the real
> noise level is flat across the FFT width, which I hope is true for all
> narrowband work, but not necesssarily so for full SSB bandwidth
> sampling.
>
> Andy  G4JNT
>