In my last post I presented the circuit of the exciter unit of the "PIXET" transmitter, a companion unit for "PIXER" receiver described earlier. In this second part of it I am describing the RF speech processor unit, the second mixer and pre driver circuitry.
Basics of RF speech processing: Imagine a power amplifier designed for 10 W pep driven by a mean ssb signal, which at least will be down by 6 db below the peak. This means a minimum output power of 2.5 watts and a resulting S-meter reading of one step down from the peak. Equalizing the dynamic range of the modulating signal will result in a better effectivity of the power amplifier as this will raise the "mean" output power. Even if this might not be directly visible at the receiver S meter, the compression of the dynamic range will increase the readability and the SNR at the receiving side. In practice, it could be proven that a – moderate - clipping limit of 20 db virtually simulates an 80 watts transmitter while, in reality, the pep output is only 10 watts. Let's understand the root of this philosophy.
Basics of RF speech processing: Imagine a power amplifier designed for 10 W pep driven by a mean ssb signal, which at least will be down by 6 db below the peak. This means a minimum output power of 2.5 watts and a resulting S-meter reading of one step down from the peak. Equalizing the dynamic range of the modulating signal will result in a better effectivity of the power amplifier as this will raise the "mean" output power. Even if this might not be directly visible at the receiver S meter, the compression of the dynamic range will increase the readability and the SNR at the receiving side. In practice, it could be proven that a – moderate - clipping limit of 20 db virtually simulates an 80 watts transmitter while, in reality, the pep output is only 10 watts. Let's understand the root of this philosophy.
The Felcher-Munson Philosophy: Grokking this theory is a bit beyond my brain right now, but the Fletcher–Munson curves are one of many sets of equal-loudness contours for the human ear, determined experimentally by Harvey Fletcher and Wilden A. Munson, and reported in a 1933 paper entitled "Loudness, its definition, measurement and calculation". The first research on the topic of how the ear hears different frequencies at different levels was conducted by Fletcher and Munson in 1933. In 1937 they created the first equal-loudness curves. Until recently, it was common to see the term 'Fletcher–Munson' used to refer to equal-loudness contours generally, even though a re-determination was carried out by Robinson and Dadson in 1956, which became the basis for an ISO 226 standard.
It is now better to use the generic term
"equal-loudness contours", especially as a recent survey by ISO
redefined the curves in a new standard. According to the ISO report, the
Robinson–Dadson results were the odd one out, differing more from the current standard
than did the Fletcher Munson curves. The report states that it is fortunate
that the 40-phon Fletcher–Munson curve on which the A-weighting standard was
based turns out to have been in agreement with modern determinations. The
article also comments on the large differences apparent in the low-frequency
region, which remain unexplained. Possible explanations are:
1. The equipment used was not properly
calibrated.
2. The criteria used for judging equal
loudness at different frequencies had differed.
3. Subjects were not properly rested for
days in advance, or were exposed to loud noise in traveling to the tests which tensed the tensor tympani and stapedius
muscles controlling low-frequency mechanical coupling.
Thus equal-loudness curves derived using
headphones are valid only for the special case of what is called
side-presentation, which is not how we normally hear. Real-life sounds arrive
as planar wavefronts, if from a reasonably distant source. If the source of
sound is directly in front of the listener, then both ears receive equal
intensity, but at frequencies above about 1 kHz the sound that enters the ear
canal is partially reduced by the masking effect of the head, and also highly
dependent on reflection off the pinna (outer ear). Off-centre sounds result in
increased head masking at one ear, and subtle changes in the effect of the
pinna, especially at the other ear. This combined effect of head-masking and
pinna reflection is quantified in a set of curves in three-dimensional space
referred to as head-related transfer functions (HRTFs). Frontal presentation is
now regarded as preferable when deriving equal-loudness contours, and the
latest ISO standard is specifically based on frontal and central presentation.
The A-weighting curve—in widespread use
for noise measurement—is said to have been based on the 40-phon Fletcher–Munson
curve. However, research in the 1960s demonstrated that determinations of
equal-loudness made using pure tones are not directly relevant to our
perception of noise. This is because the cochlea in our inner ear analyzes
sounds in terms of spectral content, each "hair-cell" responding to a
narrow band of frequencies known as a critical band. The high-frequency bands
are wider in absolute terms than the low frequency bands, and therefore
"collect" proportionately more power from a noise source. However,
when more than one critical band is stimulated, the outputs of the brain sum
the various bands to produce an impression of loudness. For these reasons
Equal-loudness curves derived using noise bands show an upwards tilt above 1
kHz and a downward tilt below 1 kHz when compared to the curves derived using
pure tones.
BBC Research conducted listening trials in
an attempt to find the best weighting curve and rectifier combination for use
when measuring noise in broadcast equipment, examining the various new
weighting curves in the context of noise rather than tones, confirming that
they were much more valid than A-weighting when attempting to measure the
subjective loudness of noise. This work also investigated the response of human
hearing to tone-bursts, clicks, pink noise and a variety of other sounds that,
because of their brief impulsive nature, do not give the ear and brain
sufficient time to respond.
What does that actually mean: The way to read this graph is as follows: look at the blue curve at the 1 kHz / 40 dB point. Now follow the curve towards the left until you reach 50 Hz on the horizontal axis. You should now read about 70 dB on the vertical axis. In essence, this states that in order for a 50 Hz tone to be perceived as loud as a 1 kHz tone is at 40 dB, it needs to be played at 70 dB. That’s 30 dB difference! A similar thing happens when you move into the high frequencies. A 10 kHz tone needs to be played at about 55 dB to be perceived at the same loudness level. Notice that this difference in loudness evens out as the volume increases (the curves higher up in the figure), for example at 100 dB, the curves have flatten out considerably, meaning the perceived loudness difference between tones at different frequencies decreases. There are two important things to take away from these curves:
1. We are less sensitive to low and high frequencies, we hear mid frequencies more prominently (especially between 1-5 kHz)
2. As the volume increases, this perceived loudness difference between the frequencies diminishes.
However, this made the basis of one of the pioneering developments in low power DX voice communication in which the high amplitude vocals are compressed for an even distribution of power over the usable bandwidth. Based upon this research; in HF-SSB radio technology in the era of late sixties, became a dependable method of modifying the speech waveform in the transmitter to produce a marked improvement in the signal-to-noise ratio at the receiver without also causing any significant increase in distortion products, either in-band or out-of-band. Since RF speech processing was the key to the performance of low-power HF-SSB radio sets - and is now recognized almost as a sine qua non in SSB transmitters - the principles involved will be described briefly. Typically, unprocessed speech has a ratio of instantaneous peak to average power of about 16dB (see Pictures below:
Unprocessed Signal
Processed Signal
In a peak-power-limited system, such as an
SSB transmitter, this represents a considerable loss of potential output power,
so some means of compressing the dynamic range of the speech signal is required
before transmission. It is now well known that clipping (or hard limiting) the
peaks of an SSB waveform, and then filtering by a second bandpass filter
similar to that in a filter-type SSB generator to remove the resulting harmonic
and high-order products, can markedly improve the articulation index of the transmitted
signal. Methods of doing this were just being developed around mid sixties.
Though modern day transmitters implement compression through DSP techniques
using digital algorithms but this project describes a very elegant alternative
compressor design like one of the yore.
PIXET Speech processor: The following schematic illustrates
the complete circuit diagram of the speech processor and the
second X-tal filter:
The SSB signal from the exciter unit is compressed using diode D1 and the base collector junction of transistor Q1. While developing the clipper circuit I initially employed two back to back diodes. But at such a low level signals adequate level of clipping demands for special hot carrier diodes like HP 5082-2811; which are both expensive and hard to find for an average experimenter. Consequently I zeroed my choice for this simple and effective alternative circuit. Both threshold and gain controls are required to be adjusted carefully. An oscilloscope is quite invaluable tool to do this precious adjustment but in case of its non- availability on the air adjustment also provides convincing results. You can feed the output signal of the transmitter into a dummy load and the adjustments of required gain and proper compression can be done by hearing the signal in a nearby receiver.
Second mixer and pre-driver: There is nothing special to explain in this section. An SBL-1 mixer is used but a home brew variety of double balanced diode mixer will perform equally well. Attempts are made to terminate all mixer ports towel defined 50 ohms impedance to ensure optimum IMD performance. The band pass filter is W7ZOI design which can be scaled to other bands of interest if desired. I tested the transmitter on 14MHz band by feeding VFO signal from 18.43 to 18.78 Mhz. The higher side VFO injection automatically puts you on the right side of sideband i.e. USB. For lower sideband operation as on 80 and 40 meters, lower side VFO injection could be used. Or you can just put up a two banded with single IF just by band switching VFO signal. In that case sideband selection would be automatic for both bands, but the VFO tuning will be in opposite direction. This will help a newbie to assemble a two bander with minimum effort and cost.
In the next post I will describe the R.F. linear amplifier for the "PIXET" transmitter.
