Hi Diego,
In this post, I will include an entire speech filterSystem except the clipper and the final undershooting final low pass anti aliasing filter.
In this correspondence, Jim makes recommendations about the lineup necessary
for an optimally designed speech system using analog design.
Note in Jim’s list of the order of circuits, he lists his 4khz undershooting filter.
Jim developed a 5 pole filter that has the same response as my pole filter, bit I prefer the 7 pole because it has much steeper skirts. You can run both filters on his software for comparison. MB
Again, I will copy directly from an email to Jim Tonne.
MB
Email :August 08/2021
Is This a complete Speech Processing system?
Hi Jim, I answer in context:
Sent from my iPhone
On Aug 8, 2021, at 12:55 PM, JimTonne <
tonne@comcast.net> wrote:
Mike:
Put the allpass stuff prior to the first AGC. This will reduce the AGC activity
and so enhance loudness. That AGC should have an attack time of the
order of 2 (absolute minimum) to 8 or so (maximum). Shorter attack times
will reduce the loudness. Longer will make the sound "pop" and is notgood-sounding.
Ok, but I do not know the term “order of 2” so what is that?
But notice I said make the attack time finite. That will allow transients to go
on through that block. Those transients will cause overmodulation. So after
the AGC you should install a clipper, even if it only works on transients.
Ok- got it.
After the clipper you should install the lowpass filter. That filter must not have
ringing in its transient response. Your filter, the 7-poler, does a very very nice
job. I ran the complete system (trapezoid generator followed by your lowpass
followed by a PWM generator followed by the output lowpass). As I varied the
frequency of the trapezoid (i.e., the clipped sinusoid) I noted the level of
modulation. Your filter has full modulation without any peaks going into
overmodulation, right up to 4 kHz. Then the modulation level rolls off. See the
plot below.
Good news! We have a saying- even a blind sow finds an acorn (or truffle ) from time to time… My hat size is not changing just yet.
Questions abound but grateful to get anything right because I define that as progress…😉
<000743.jpg>
My design uses 2 instead of 3 opamps and also has full modulation on trazezoidal
waves up to 3500 Hz instead of your 4000 Hz.
For SSB, your filter is better- I was thinking “good” quality AM here.
Might build a separate speech processor for my SSB rigs later.
I have a little circuit I use to make lopsided speech waveforms. I fired it up and ran
its lopsided waveform into the W3AM allpass. Here is the schematic of that entire
system:
<000751.jpg>
On the left is the speech simulator. On the right is the W3AM allpass.
Here is the waveform out of the speech simulator, along with entire schematic:
<000749.jpg>
The red line down the middle is zero volts so you can see how lopsided that waveform is.
Now the output of the allpass:
Yes- I am familiar with this- having read your All Pass filter paper and also your new approach to speech processing paper.
If you still have my previous email showin blocks for the speech system, please confirm that I got it right.
1. Your mic amp with treble boost snd lpf
2. W3AM All Filter
3. Your AGC-Compressor circuit
4. Your clipper
5 My version of your negative transient 4khxz lpf driver
6. My PWM modulator
<000750.jpg>
See how that allpass made the signal have equal peak values positive and negative.
Put the treble boost prior to the AGC. The allpass can be before or after the treble boost
but prior to the AGC.
- JimT
I feel like I am beginning to get a little clue about this and understand better how little I know on the subject.
I am very grateful for your wise counsel and patience.
73, MB
