Hi all
I'm so happy. That's why I am writing this. CESSB works! I
have recorded a 1-minute YouTube demo (Churchill's speech
still) showing a QMX receiving QMX+ CESSB transmissions, more
on this below.?
CESSB (Controlled Envelope Single Sideband) reminder: the
place to read about it is the original ARRL QEX article
published in November 2014 by David L. Hershberger, W9GR:
.?
However that article is focused on generating SSB by a
conventional SSB exciter. I struggled with thinking about how
to apply that to the rather special case in QMX where we are
generating SSB by EER (Envelope Elimination and Restoration)
which is a rather different process, though also involving the
inevitable Hilbert Transform.?
In the end I decided I did not NEED to do it exactly the
same way it's done in the article: in QMX SSB via EER, we have
separated the signal into phase and amplitude components. I
realized that this fact can be used to ADVANTAGE. I developed
a completely different way of generating CESSB to the one in
the article. I guess I shall have to write down all the
details carefully to document all this.?
I am still using as my audio sample (I'll call it track
from now on), Winston Churchill's "Be Ye Men of Valour" speech
from 1940 ?
Note that for these experiments I am running the QMX+
powered with 12V but with the gain set such that the RF output
is 1.0W PEP. When using full power about a 6dB improvement in
dynamic range would be expected.
In my attached photograph "CESSB Off.jpg" I photographed
the same first syllable in which he says "I". The audio
recording has an amplitude which does not exceed the range?+/-
1; in other words it reaches full scale but it does not exceed
it. Yet you can clearly see that at RF the envelope overshoots
"full scale", the dotted blue cursor lines on the 'scope
screen, which are determined using a steady 1kHz test tone. In
this case the overshoot is 32%, or 2.4 dB. You would need to
turn down your gain by 2.4 dB in order to avoid clipping and
splatter (or, if using an amplifier with ALC, the ALC would do
that for you).?
"CESSB On.jpg" I photographed the exact same first syllable
but now CESSB is switched ON. Peaks of the RF envelope which
do not exceed "full scale" indicated by the dotted blue lines
are left alone, they do not trigger the CESSB algorithm. But a
peak which exceeds the allowed maximum envelope appears
cleanly reduced without altering its shape.?
The best thing about my new method for generating CESSB is
that it is inherently clean (does not increase bandwidth) and
I do not need to run any subsequent filtering, so there is no
need for a computationally expensive linear phase FIR filter.
I am constantly reminded that although I chose a powerful
processor for QMX, a 32-bit ARM Cortex M4 at 168 MHz, it is
certainly NOT anywhere near the kind of powerful CPU that is
being used in the very few top-end amateur radio manufacturers
where CESSB is being used (Elecraft K4, Flex SDR, Apache
Anan). So when doing DSP it is critical to keep remembering
that CPU cycles are still very precious. It was therefore all
the more satisfying, finding a way to do this that involves
only a few simple arithmetic operations per DSP audio sample.?
For testing the SSB reception I used an 80-20m QMX having
the current latest production firmware version 1_00_027, in
"digi" mode, which has a filter bandwidth? of 150-3200 Hz.
This is wider than the SSB filter bandwidth used in the QMX
SSB firmware which is 300-2800 Hz. I connected the BNC RF port
of my QMX+ (the transmitter) via 80dB of inline 50-ohm BNC
attenuators to the BNC RF port of my 80-20m QMX (the
receiver). Both the QMX+ (transmitter) and QMX (receiver) are
connected to my PC and appear as USB Sound cards there. I
played the Churchill quote on VLC Media Player to the QMX+
(transmitter); and recorded the QMX (receiver) using Audacity
software.?
I have two versions of the audio clip. One of them is the
original one, scaled so that its peaks hit (but do not exceed)
full scale audio resolution. The other one was amplified by
3dB (voltage) using Audacity, allowing clipping at full scale
audio. The clipping therefore introduces harmonic content, and
increases the average power level by 3dB. In the attached
chart these are labelled #1 and #2, being the original and 3dB
amplified versions respectively.?
I used the Audacity Spectrum analysis function (48ksps,
8192 point, Hanning window) exporting to a CSV and loading it
into the spreadsheet for graphing. In the QMX Lab screen
(attached "lab.png") I can switch on or off CESSB and also
measure the peak and average envelope amplitude over a 2.4
million sample duration (200 seconds).?
The attached chart "spectrum.png" shows all four spectrum
lines. Audio track #1 is colored blue (CESSB off) and purple
(CESSB on); track #2 is colored red (CESSB Off) and orange
(CESSB On). You can see that there is about a 3dB difference
between audio tracks #1 and #2 which is expected because the
gain was increased on track #2. In each track, there is NO
discernible difference in spectrum at all, between whether
CESSB is switched Off or On.?
Therefore QMX CESSB is, as it should be, limiting the
envelope without causing any distortion and without widening
the bandwidth. It simply increases the average to peak power
without clipping or splatter or distortion.?
In my track #1 the envelope overshoot is 29% corresponding
to 2.2dB. In track #2 the envelope overshoot is 65% which is
4.3dB. This is also expected (see the CESSB article) because
when audio is amplified and clips, at the clipping point the
waveform is somewhat trapezoidal which starts to approach a
squarewave. As the article explains, squarewaves behave rather
badly in the Hilbert transform: the zeroes of a series of
squarewave harmonics line up, but when shifted 90-degrees (the
Hilbert transform) then the PEAKS all line up at the same
time. Which is the whole yuckology of the SSB situation.?
So the use of CESSB provides an increase in average power,
the exact amount depends on the nature of the speech being
transmitted; in my two example tracks the benefit was 2.2dB
and 4.3dB. Taking the latter number for example, it means that
a 5W transmitter has the average power equivalent to a 13W
transmitter, yet without adding any splatter or distortion.?
Here is my 1 minute YouTube demo video:
The SSB sounds pretty good, if you compare it to the
original Churchill speech you will certainly say it sounds
rather similar. Note that the video and audio of the video
aren't synchronized; I took the video then I overlayed the
sound from the MP3 file I had recorded on my PC. So don't
worry that what's shown on the 'scope doesn't quite match the
sound.?
CESSB is not exactly the same thing as compression or
pre-emphasis though you could say they are all quite closely
related. I am aware too that low frequency tones in SSB don't
do much for intelligibility but DO waste a lot of power
unnecessarily so the advice for well set-up SSB is to reduce
the bass tones somewhat. So compression and tone adjustments
to what comes out of an actual microphone, are still on the
ToDo list.?
