[aprssig] A Closer Look at the WA8LMF TNC Test CD

Rob Riggs mobilinkd at gmail.com
Sun Jan 3 11:53:31 CST 2016


Hi Stephen,

I am sure that is true.  The audio engineering on the test CD is top
notch.  I am grateful for the effort you have put into them.  We are all
better off for having them.

What I have found is that different model radios "hear" differently.  And
it is the output of imperfect signals, especially over-deviated or
off-frequency signals, where those differences manifest most.  I would
assume it is these sorts of packets, among others, that John is attempting
to catalog.

I have found that it is possible to over-fit an AFSK demodulator to the
test CD to the point that actual performance, when tested across multiple
radios, begins to decrease.  This is not unlike an AI training exercise,
where it is possible to over-fit the solution to the training set.

What I have found works best for me is to have training samples taken from
multiple sources to tune and validate a demodulator.

I agree with Scott that it would be good to have a larger sample set to
work with.  With the advent of the "ultra-cheap SDR" it should be possible
to enlist hams to collect these directly from across the globe.



Kind Regards,

Rob Riggs WX9O
Mobilinkd LLC

On Sat, Jan 2, 2016 at 11:01 AM, Stephen H. Smith <wa8lmf2 at aol.com> wrote:

> On 1/2/2016 2:04 AM, Rob Riggs via aprssig wrote:
>
>> Hi John,
>>
>> Some of the distortion you show, especially the ones Dire Wolf shows the
>> volume
>> level over 100, look a lot like what one sees when overdriving the inputs
>> on an
>> audio capture device.
>>
>> How much of the distortion you are seeing is caused by the capture and
>> recording process?  How much is due to the radio used to capture the
>> signal?
>> Those seem to be unknowns and ought to be accounted for.
>>
>>
> I made a serious effort to eliminate these kinds of issues when making the
> recordings.
>
> I did the original capture with a Soundblaster PCI 128 card (the former
> Ensoniq) and pointedly kept the peak level at -6 dB from the max possible
> in Audition.   During the subsequent editing, I used Audition to normalize
> the audio level to within .5 dB of max; i.e each clip was scanned and
> adjusted so that the loudest point in the clip was .5 dB down from the max
> possible. (This is a routine process in digital audio editors.)
>
> Since the un-squelched un-quieted receiver white noise between
> transmissions was always higher than even fully-deviated signals,
> peak-level overload and clipping would have only occurred on the noise, not
> the signals anyway.
>
> The other issue would be the noise floor and distortion of the sound card
> itself. I initially tested the card's performance by recording a single
> 1000 Hz tone from one of my IFR-1500 service monitor's audio generators. I
> then played the tone back through the playback side of the card into the
> IFR's SINAD meter.  In addition to the usual reverse-scale analog SINAD
> meter display that makes it hard to read more than about a 20dB SINAD, the
> IFR can produce a digital readout of SINAD on it's scope screen that will
> read out down to the noise floor of the device at about 45 dB.
>
> The result of the "round trip recording" was a 30+ dB SINAD, indicating
> that the total noise and distortion added by a round trip through the sound
> card card was less than 0.1%.
>
> To verify the cleanliness of the audio source, I fed the 1K tone from the
> IFR's audio generator directly into it's own SINAD meter input.  The meter
> showed only the residual noise floor of around 45 dB SINAD.
>
> To eliminate the possibility of AC power-supply ground loops
> "contaminating" the measurements, I used a 12 VDC battery to power the IFR
> using it's DC input during the tests.
>
>
> The radio used for the off-the-air tests was a Yaesu FT-1500 compact 2M FM
> monobander tapped through it's 6-port mini-DIN "data port" using the "9600
> baud" pin (i.e. the connection directly to the radio's raw discriminator
> output).  A test with the IFR's RF generator set to 50 microvolts and
> modulated to 3.5K deviation with the 1K audio tone produced a 38 dB SINAD
> at the data port pin.  (And only about 21 dB at the speaker terminal,
> showing the contribution of the less-than-Hf-Fi audio power amp!) The
> din-port SINAD went down to about 21 dB when I advanced the generator
> deviation to a full 5 KHz.
>
>
> Bottom line: Any distortion you are seeing is most likely due to either
> the original on-the-air signal being distorted, or the playback CD player
> being used having mediocre D/A converters or output line driver amps.
>
> ______________________________________________________________________
>
>
> --
>
> Stephen H. Smith    wa8lmf (at) aol.com
> Skype:        WA8LMF
> EchoLink:  Node #  14400  [Think bottom of the 2-meter band]
> Home Page:          http://wa8lmf.net
>
>  _______ Windows 10 Outrages! _______
>    <http://WA8LMF.net/Windows10_Info>
>
> Live Off-The-Air APRS Activity Maps
>    <http://wa8lmf.net/map>
>
> Long-Range APRS on 30 Meters HF
>    <http://wa8lmf.net/aprs/HF_APRS_Notes.htm>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.tapr.org/pipermail/aprssig/attachments/20160103/0a2a7d5d/attachment.html>


More information about the aprssig mailing list