From: Paul
Date: Fri, 29 Mar 2002 09:14:04 +0000
Subject: Re: Ready :-)) - Re: [TSSP] short H/D and stuff
Terry wrote:
> I hooked up the big coil and the small coil to the pinger at the
> same time and took the following data files. They are all
> zipped too at:
> http://hot-streamer.com/tssp/3-28/
The FT is in
http://www.abelian.demon.co.uk/tssp/tcma/ft3.gif
extracted from the avg-128 trace.
TEK00007.CSV
PK FREQ kHz (Error +/-) Q FACTOR (Error +/-) LEVEL Indent
1 351.208 (0.01%,29Hz) 216.04 (1.65%, 3.6) -2.0dB A-f1/4
2 143.476 (0.01%,12Hz) 241.10 (0.11%, 0.3) -4.9dB B-f1/4
3 873.002 (0.03%,239Hz) 243.24 (0.45%, 1.1) -16.7dB A-f3/4
4 511.845 (0.12%,591Hz) 112.90 (1.43%, 1.6) -23.8dB B-f5/4
5 1265.541 (0.21%,2599Hz) 191.11 (3.92%, 7.5) -22.8dB A-f5/4
6 666.399 (0.12%,770Hz) 91.25 (1.65%, 1.5) -27.7dB B-f7/4
7 1602.539 (0.49%,7804Hz) 138.07 (6.98%, 9.6) -26.1dB A-f7/4
8 819.766 (1.15%,9467Hz) 78.88 (6.98%, 5.5) -28.4dB B-f9/4
9 977.371 (1.15%,11287Hz) 67.92 (9.31%, 6.3) -32.8dB B-f11/4
10 1133.085 (1.15%,13085Hz) 61.56 (16.55%,10.2) -33.3dB B-f13/4
Accounted for 93.63% of input signal
Coil A is tfsm1: bare d=0.108m h/d=6.14 sr=0.91 b/h=0.03 turns=1176
Coil B is tfltr: bare d=0.261m h/d=2.92 sr=0.67 b/h=0.03 turns=1000
Coil B's 3/4 wave mode at around 353 kHz is completely submerged
under the fundamental resonance of coil A at 351 kHz. This is
causing the anomalous high error on PK1 Q factor - the extra bit
of signal lurking underneath is preventing full convergence by tcma.
Apart from this collision (which would prevent useful readings
of either A-f1/4 or B-f3/4), the pinging of two coils is working
perfectly. The frequencies line up nicely with tssp model
predictions for the two coils, and this allows unambiguous
identification of the peaks. There is no sign of one coil
affecting the other, but it would be interesting to do two
successive runs, one with the coils in present position, and
another with them further apart.
There is a small but worthwhile improvement between the avg-16
and the avg-128 traces, so I think we should use 128 always in
future. This little improvement doesn't seem to increase
accuracy of Q (that's limited by the Y quantisation), but it does
allow tcma to tackle weaker peaks.
You'll notice a trend towards greater Q error as the mode amplitude
decreases. This is due to two factors:
a) The weaker signals are represented by a smaller number of
vertical resolution steps, due to their lower amplitude wrt to the
scope's dynamic range.
b) The Q factors tend to be smaller, so that waveform matching
occurs over fewer horizontal time steps.
Mostly the limitation is (a). This could be obviated to some
extent by pre-emphasis between the current probe output and the
scope input terminals. Ideally the signal should be differentiated
before going into the scope, to give a 6dB/octave rise. That way
the higher overtone components will have increased amplitude
without the fundamental reaching the limits of the Y amp. The
price of this is increased noise across the board, thus reducing
accuracy a little for the stronger signals.
The theoretical accuracy on Q (for a noise free signal) is
roughly
100/sqrt(NY * NT) percent
where NY is the number of vertical resolution steps occupied by
the signal, and NT is the number of time steps in the trace between
the start of the ping, and the point where it has decayed to 10%.
For f1, typically NY = 256 (using half the Y range to prevent
limiting) and NT = 6000, so Error(f1) should be around 0.08% at
best. For a 3/4 wave mode at 10% initial amplitude (typical),
we have NY = 30, and NT is around 2000, so Error(f3) is at best
0.4%. At f5, we have NY circa 4, NT circa 1000, Error(f5) at best
is 1.6%. Mileage varies since NT depends on the mode Q, and NY
on the Y gain settings. Thus it's important to try to use the
max dynamic range of the scope, but at the same time it's vital to
avoid limiting in the Y-amp. Flat tops will really mess up the
algorithm! I might add a bit of code into tcma to detect and warn
of flat tops.
--
Paul Nicholson,
--
Maintainer Paul Nicholson, paul@abelian.demon.co.uk.