Introduction
Being into weak signal work, either in contests or DXing, I fight against noise.

Europe is noisier than other parts of the world. In The Netherlands man-made noise
is becoming a significant issue, even a problem. However, with the right attention it is possible
to reduce noise/interference and become competitive on shortwave, even from The Netherlands.

Many people know me advocating the (vertically polarized) deltaloop on 40m. When properly fed
this antenna behaves like two stacked full size verticals but .  .  .  moreover, it reduces noise
during reception because it’s a closed loop/system. “If you can’t hear them, you can’t work them!”

In my quest reducing noise I was confronted with a kind of ‘nuisance after discovery’: audio noise.

My good old buddy, a FT-857 (yes . . . not a 857D), always assists me during holiday trips to
make contacts from exotic places. A few weeks ago I was in such an exotic place and was confronted
with a HUGE pile-up on 40m and tried to pull out weak JA’s and W’s out of the noise with my
simple setup: a (heavily modded) barefoot FT-857 (60 – 65W o/p) and 40m (vert. pol.) deltaloop in CW.

Blame my 70, 23, 13, 9, 6, 3cm, 2, 6, 40, 80 and 160m experience, but even on short wave you’ve weak signals!
Because I listened so much into the noise, my brains developed a kind of ‘weak signal filter’ in pile-ups.
During the last operation from this exotic place something became apparent to me which I (apparently)
didn’t notice before: unwanted audio noise in my (LoFi, NOT HiFi!) headphones.

Back in The Netherlands I decided to dig further into this issue.

Observation
Back home I inserted a weak signal (-140 dBm @7025.0 kHz) into my FT-857 and listened through my
headphones. While listening in CW-N (300 Hz BW) I was fighting against the perception ‘do I hear a tone or not?’.

In order to rule out some bias in myself I asked someone else to push the RF-on/off button of my
signal generator. My score was around 60% correctly detecting the tone or not.

With this applied weak signal I discovered I had problems in determining whether I could hear a tone
(or not) was related to the volume but increasing the volume also increased high(er) frequency noise
(more?) in the audio! Certainly a psycho acoustic issue is involved.

In other words, the high(er) frequency noise was interfering with my perception (whether there was
a tone . . . or not). In other other words, this high frequency noise was an extra load in filtering
the tone (in my brains?). The high frequency noise can be described as a ‘high pitched hiss’.

This led me to investigate the audio receive chain of the FT-857 and soon I focussed on the TDA2003
audio amplifier. Simply because this stage has an almost equivalent gain contribution in the whole
chain than it’s predecessing (RF and IF) stages. Assume that the -140 dBm signal has to be converted in
1W audio output (+30 dBm), the whole chain must have an amplication of 140 + 30 = 170 dB.

Knowing that generic audio amplifiers have (let’s say) 80 – 90 (if not 100) dB gain,
it’s relative contribution has might be significant (not for the overall noise figure (NF)
but more the timbre of the sound).

So . . . I decided to dig into the datasheet of the TDA2003 audio amplifier.

TDA2003
The TDA2003 is a standard audio amplifier. On the first page of the datasheet I found the standard application:

Immediately ‘Rx’ and ‘Cx’ drew my attention. Further in the datasheet it is explained that Rx and Cx
introduce a low pass behaviour in the above circuit.

In the FT-857 circuit diagram I noticed Cx and Rx were lacking (see below):

From the formulae given in the original datasheet, and assuming ‘B’ in the formula for Cx is
the cut off frequency, I calculated Rx = 100 Ohm, and Cx around 220 nF for B = 3000 Hz.
Because I hadn’t 220 nF and fiddling with  2x 100 nF parallel wasn’t my cup of tea, I chose 470 nF,
also to be on ‘the safe side’. Thus, 470 nF in series with 100 Ohm were connected between
legs 2 and 4 of the TDA2003.

Measurements
My first impression by ear was that the mod works, in a sense that I was more able to detect the tone.
Again someone else was asked to randomly press the RF-on/off button of my signal generator and
my score was 100% correct in determining whether there was a tone, or not.

We did a second test where I had to leave my shack. After I was called I had to take place at the table,
put on my headphones and the above procedure was repeated. At first glance I didn’t hear anything
but noise, but suddenly I thought I heard something. However, this seemed weaker than the previous tests.

I turned up the volume somewhat and after my brains ‘locked’ into the situation I was able to score
around 70%. It seemed my assistant reduced the signal generator level to -146 dBm by inserting a calibrated
20 dB attenuator and adjusting the output of the signal generator to -126 dBm !

Of course I was curious if I could determine the yield of the mod in a more objective manner.
A few years ago I used the ‘Analyzer2000′ program to determine and optimise signal to noise ratios (SNR).
However, it seems that this program doesn’t run anymore on 64 bit Windows systems. As alternative I installed
Spectrum Lab from DL4YHF. Despite Spectrum Lab is advertised being able to determine SNR’s, I couldn’t
find a separate SNR option or button at first glance. This program is significantly more versatile and
complicated than Analyzer2000, so I reckon I have to dig further in the manual ; -).

Anyway, as alternative screenshots of the waterfalls were saved in order to see the difference in noise
performance. ‘Before and after mod’ screenshots are saved into the animated gif depicted below.
In this animated gif three filter responses are displayed from top to bottom:
CW (CFIL, i.e. original filter), CW (2300 Hz BW, ‘SSB-filter’), and CW-N (300 Hz BW).

The crispy yellow points are the FT-857 audio beeps when selecting another filter.

Conditions: signal of -140 dBm @7025.0 kHz inserted into the antenna connector of the FT-857.
Audio extracted from the front headphone connector with headphones as parallel/speaker load.

It can be seen that the screenshot taken at around 19.48 hrs (‘after mod’) is more dark/black in the
higher frequency region than the 19.36 one. Of course this method is very ‘wicky whacky’ but it gives
some visual information about the audio (power) distribution. There may be some slight volume differences
between the two screenshots, due to implementation of the mod . . .

Anyway, the mod works for my brains, may increase (perceived) SNR with around +6dB as determined with a
‘wicky whacky’ method, and . . . is easy to realize!  See picture below (click to enlarge in new tab).