a little oscillator theory

[John Larkin]

Among other parasitic functions, I'm the local LC oscillator designer.

A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.

I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.

Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.

But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.

Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

[John Larkin]

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

A violin uses a negative resistance source (the bow) and a resonator
with multiple modes.

So does a human voice. I recall that Joan Baez could make almost
perfect sine waves.

[Phil Hobbs]

Post by John Larkin

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

A violin uses a negative resistance source (the bow) and a resonator
with multiple modes.
So does a human voice. I recall that Joan Baez could make almost
perfect sine waves.

Which contain zero information, naturally. ;)

Cheers

Phil Hobbs
(Now if _I_ had a hammer....)

[Jeroen Belleman]

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

I read somewhere that to get the lowest possible phase noise,
the gain element should be cut off for most of the cycle while
the tank freewheels. Once per period, it should give a little
kick at the peak of the cycle, where that doesn't affect the
phase. I've also seen arguments that say this is false.

I haven't tried it and I haven't analyzed it in detail. YMMV.

Touch-tone oscillators are interesting. They oscillate on two
frequencies at the same time. Squegging is interesting too:
It's put to good use in super-regenerative receivers.

Jeroen Belleman

[John Larkin]

On Fri, 23 Feb 2024 18:29:07 +0100, Jeroen Belleman

Post by Jeroen Belleman

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

I read somewhere that to get the lowest possible phase noise,
the gain element should be cut off for most of the cycle while
the tank freewheels. Once per period, it should give a little
kick at the peak of the cycle, where that doesn't affect the
phase. I've also seen arguments that say this is false.
I haven't tried it and I haven't analyzed it in detail. YMMV.
Touch-tone oscillators are interesting. They oscillate on two
It's put to good use in super-regenerative receivers.
Jeroen Belleman

Loading Image...

I think the original touch-tone (maybe this is it) used an expensive
point-contact germanium transistor to provide a negative resistance to
two LC resonators in series. Each LC had its own varistor to limit
amplitude.

[piglet]

Post by John Larkin
On Fri, 23 Feb 2024 18:29:07 +0100, Jeroen Belleman

Post by Jeroen Belleman

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

I read somewhere that to get the lowest possible phase noise,
the gain element should be cut off for most of the cycle while
the tank freewheels. Once per period, it should give a little
kick at the peak of the cycle, where that doesn't affect the
phase. I've also seen arguments that say this is false.
I haven't tried it and I haven't analyzed it in detail. YMMV.
Touch-tone oscillators are interesting. They oscillate on two
It's put to good use in super-regenerative receivers.
Jeroen Belleman

https://www.dropbox.com/s/br6namqyxas1f02/Bell_DTMF.jpg?raw=1
I think the original touch-tone (maybe this is it) used an expensive
point-contact germanium transistor to provide a negative resistance to
two LC resonators in series. Each LC had its own varistor to limit
amplitude.

Yes the AGC is very important for dual resonance oscillators, without it
you tend to get oscillation at one or other of the two frequencies but not
reliably both together.

[Joe Gwinn]

Post by John Larkin
On Fri, 23 Feb 2024 18:29:07 +0100, Jeroen Belleman

Post by Jeroen Belleman

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

I read somewhere that to get the lowest possible phase noise,
the gain element should be cut off for most of the cycle while
the tank freewheels. Once per period, it should give a little
kick at the peak of the cycle, where that doesn't affect the
phase. I've also seen arguments that say this is false.
I haven't tried it and I haven't analyzed it in detail. YMMV.
Touch-tone oscillators are interesting. They oscillate on two
It's put to good use in super-regenerative receivers.
Jeroen Belleman

<Loading Image...
I think the original touch-tone (maybe this is it) used an expensive
point-contact germanium transistor to provide a negative resistance to
two LC resonators in series. Each LC had its own varistor to limit
amplitude.

As I recall, that is exactly how it works. The drawing above came
from the BSTJ article on the design, published in the 1950s.

It helped to rearrange the circuit diagram, simplifying the tone pad
switching details.

I have the article somewhere, from when I was designing some
oscillators in the 1970s, but have not seen it in decades.

Joe Gwinn

[Liz Tuddenham]

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.

After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.

As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

[john larkin]

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

I'm thinking that my next oscillator will use an IC, specifically a
BUF602, instead of a phemt as the gain element. It's a1 GHz unity-gain
buffer that I assume TI has done right. It shouldn't oscillate at 6
GHz or whatever.

It still needs some controlled amplitude limiting, which I'll do with
a diode. I sure don't want the BUF602 to rail to limit oscillation
amplitude.

[Jan Panteltje]

On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

[John Larkin]

Post by Jan Panteltje
On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

Nowadays one can do a DDS phase accumulator and a sine lookup, in an
FPGA or all software in a uP. Frequency and amplitude and even phases
and sweeps can be precisely programmed.

It does need a pretty good lowpass filter after the DAC to take out
the jaggies.

[Jan Panteltje]

On a sunny day (Sat, 24 Feb 2024 11:41:28 +0000) it happened

Post by Liz Tuddenham

Post by Jan Panteltje
On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

What was the amplitude variation and noise after filtering? I needed at
least 96dB below signal and a few stages of filtering could actually
generate more noise than it removes.

The amplitude variation over the audio range was next to zero (DAC output),
the filtering was just an RC with -3 dB point above the maximum audio I used (20 kHz IIRC).
'noise' is the number of steps I think, so for an 8 bit sinewave you had 256 steps
If you use 12 bits you get more smaller steps.
You could make a RC filter with opamps, then anything is possible.
I build that because I needed to test some systems, quick hack...

An other way to make a changing audio frequency is mix 2 RF signals,
one from a fixed crystal oscillator, the other maybe an LC oscillator if you need a big range
or also an xtal oscillator if you need a narrow range.
Most xtal oscillators are tunable over some range, I use that all the time,
Say if you mix 10 MHz with 10.1 MHz you get 100 kHz
Filtering out the 10+ MHz is then easy.
Tuning is also easy, varicap, or some transistor used as varicap.
2 RF signals, 1 MHz and 1.01 MHz on your AM radio will demonstrate that (10 kHz out).
96 dB is a lot, best of luck!

[Liz Tuddenham]

Post by Jan Panteltje
On a sunny day (Sat, 24 Feb 2024 11:41:28 +0000) it happened

Post by Liz Tuddenham

Post by Jan Panteltje
On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

What was the amplitude variation and noise after filtering? I needed at
least 96dB below signal and a few stages of filtering could actually
generate more noise than it removes.

The amplitude variation over the audio range was next to zero (DAC
output), the filtering was just an RC with -3 dB point above the maximum
audio I used (20 kHz IIRC). 'noise' is the number of steps I think, so for
an 8 bit sinewave you had 256 steps If you use 12 bits you get more
smaller steps. You could make a RC filter with opamps, then anything is
possible. I build that because I needed to test some systems, quick
hack...
An other way to make a changing audio frequency is mix 2 RF signals, one
from a fixed crystal oscillator, the other maybe an LC oscillator if you
need a big range or also an xtal oscillator if you need a narrow range.
Most xtal oscillators are tunable over some range, I use that all the
time, Say if you mix 10 MHz with 10.1 MHz you get 100 kHz Filtering out
the 10+ MHz is then easy. Tuning is also easy, varicap, or some transistor
used as varicap. 2 RF signals, 1 MHz and 1.01 MHz on your AM radio will
demonstrate that (10 kHz out). 96 dB is a lot, best of luck!

Yes, it took several re-designs to get 96dB, but I did eventually
achieve it. Even potentiometers are too noisy for level control at
those sorts of S/N ratios, so well-smoothed DC control of the oscillator
amplitude was the only way. The gain-setting pot on the detector was
too noisy for the most sensitive measurements, so it had to be left at
one end of its travel and the oscillator output used to control the
signal level instead.

It did mean that I found a completely unexpected extra use for the
intermodulation meter: it could be used to test potentiometers and
variable resistors for self-generated noise.

[John Larkin]

Post by Jan Panteltje
On a sunny day (Sat, 24 Feb 2024 11:41:28 +0000) it happened

Post by Liz Tuddenham

Post by Jan Panteltje
On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

What was the amplitude variation and noise after filtering? I needed at
least 96dB below signal and a few stages of filtering could actually
generate more noise than it removes.

The amplitude variation over the audio range was next to zero (DAC output),
the filtering was just an RC with -3 dB point above the maximum audio I used (20 kHz IIRC).
'noise' is the number of steps I think, so for an 8 bit sinewave you had 256 steps
If you use 12 bits you get more smaller steps.
You could make a RC filter with opamps, then anything is possible.
I build that because I needed to test some systems, quick hack...
An other way to make a changing audio frequency is mix 2 RF signals,
one from a fixed crystal oscillator, the other maybe an LC oscillator if you need a big range
or also an xtal oscillator if you need a narrow range.
Most xtal oscillators are tunable over some range, I use that all the time,
Say if you mix 10 MHz with 10.1 MHz you get 100 kHz
Filtering out the 10+ MHz is then easy.
Tuning is also easy, varicap, or some transistor used as varicap.
2 RF signals, 1 MHz and 1.01 MHz on your AM radio will demonstrate that (10 kHz out).
96 dB is a lot, best of luck!

In your circuit, every cycle has the exact same DAC steps. In a DDS
system, the dac steps squirm from cycle to cycle. Noise-wise, DDS is
about as good, provided it has an ideal lowpass filter after the DAC.

With a decent ratio of clock to sine frequency, say 20:1, the lowpass
isn't terribly hard. But the DDS will always be noisier than the
sychronous scheme.

A combo would be best: make a DDS frequency synthesizer, a phase
accumulator, sine lookup, DAC, lowpass filter, comparator. Then use
that to clock a counter, sine lookup, DAC, filter. That just takes a
few more parts and some code. Share the sine lookup.

Possibly eliminate the first DAC, or make a 2-bit DAC with resistors.

[Jeroen Belleman]

Post by John Larkin

Post by Jan Panteltje
On a sunny day (Sat, 24 Feb 2024 11:41:28 +0000) it happened

Post by Liz Tuddenham

Post by Jan Panteltje
On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

What was the amplitude variation and noise after filtering? I needed at
least 96dB below signal and a few stages of filtering could actually
generate more noise than it removes.

The amplitude variation over the audio range was next to zero (DAC output),
the filtering was just an RC with -3 dB point above the maximum audio I used (20 kHz IIRC).
'noise' is the number of steps I think, so for an 8 bit sinewave you had 256 steps
If you use 12 bits you get more smaller steps.
You could make a RC filter with opamps, then anything is possible.
I build that because I needed to test some systems, quick hack...
An other way to make a changing audio frequency is mix 2 RF signals,
one from a fixed crystal oscillator, the other maybe an LC oscillator if you need a big range
or also an xtal oscillator if you need a narrow range.
Most xtal oscillators are tunable over some range, I use that all the time,
Say if you mix 10 MHz with 10.1 MHz you get 100 kHz
Filtering out the 10+ MHz is then easy.
Tuning is also easy, varicap, or some transistor used as varicap.
2 RF signals, 1 MHz and 1.01 MHz on your AM radio will demonstrate that (10 kHz out).
96 dB is a lot, best of luck!

In your circuit, every cycle has the exact same DAC steps. In a DDS
system, the dac steps squirm from cycle to cycle. Noise-wise, DDS is
about as good, provided it has an ideal lowpass filter after the DAC.
With a decent ratio of clock to sine frequency, say 20:1, the lowpass
isn't terribly hard. But the DDS will always be noisier than the
sychronous scheme.
A combo would be best: make a DDS frequency synthesizer, a phase
accumulator, sine lookup, DAC, lowpass filter, comparator. Then use
that to clock a counter, sine lookup, DAC, filter. That just takes a
few more parts and some code. Share the sine lookup.
Possibly eliminate the first DAC, or make a 2-bit DAC with resistors.

I was sort-of shocked when I came across a project with a local
oscillator for a spectrum/network analyzer where the designer
had used a DDS to provide the reference frequency for a PLL.
The DDS was used just to cover the interval between two
successive steps of the PLL. Quite over the top, I'd say.

Then again, frequency synthesizers with small steps can get
complicated.

Jeroen Belleman
the PLL.

[John Larkin]

On Sat, 24 Feb 2024 22:42:22 +0100, Jeroen Belleman

Post by Jeroen Belleman

Post by John Larkin

Post by Jan Panteltje
On a sunny day (Sat, 24 Feb 2024 11:41:28 +0000) it happened

Post by Liz Tuddenham

Post by Jan Panteltje
On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

What was the amplitude variation and noise after filtering? I needed at
least 96dB below signal and a few stages of filtering could actually
generate more noise than it removes.

The amplitude variation over the audio range was next to zero (DAC output),
the filtering was just an RC with -3 dB point above the maximum audio I used (20 kHz IIRC).
'noise' is the number of steps I think, so for an 8 bit sinewave you had 256 steps
If you use 12 bits you get more smaller steps.
You could make a RC filter with opamps, then anything is possible.
I build that because I needed to test some systems, quick hack...
An other way to make a changing audio frequency is mix 2 RF signals,
one from a fixed crystal oscillator, the other maybe an LC oscillator if you need a big range
or also an xtal oscillator if you need a narrow range.
Most xtal oscillators are tunable over some range, I use that all the time,
Say if you mix 10 MHz with 10.1 MHz you get 100 kHz
Filtering out the 10+ MHz is then easy.
Tuning is also easy, varicap, or some transistor used as varicap.
2 RF signals, 1 MHz and 1.01 MHz on your AM radio will demonstrate that (10 kHz out).
96 dB is a lot, best of luck!

In your circuit, every cycle has the exact same DAC steps. In a DDS
system, the dac steps squirm from cycle to cycle. Noise-wise, DDS is
about as good, provided it has an ideal lowpass filter after the DAC.
With a decent ratio of clock to sine frequency, say 20:1, the lowpass
isn't terribly hard. But the DDS will always be noisier than the
sychronous scheme.
A combo would be best: make a DDS frequency synthesizer, a phase
accumulator, sine lookup, DAC, lowpass filter, comparator. Then use
that to clock a counter, sine lookup, DAC, filter. That just takes a
few more parts and some code. Share the sine lookup.
Possibly eliminate the first DAC, or make a 2-bit DAC with resistors.

I was sort-of shocked when I came across a project with a local
oscillator for a spectrum/network analyzer where the designer
had used a DDS to provide the reference frequency for a PLL.
The DDS was used just to cover the interval between two
successive steps of the PLL. Quite over the top, I'd say.
Then again, frequency synthesizers with small steps can get
complicated.
Jeroen Belleman
the PLL.

The SRS CG635 clock generator works like that. A DDS drives a very
narrowband crystal bandpass filter which then becomes the reference
for a PLL. The PLL can only generate frequencies that are related to
integer divisors, so the DDS interpolates. The bandpass filter cleans
up the DDS spurs.

Yes, complicated.

The 635 is a nice box. It will generate clocks or PRBS sequences up to
2 GHz. It has the usual SRS archaic 7-segment display and abominable
user interface.

[Liz Tuddenham]

Jan Panteltje <***@comet.invalid> wrote:

[...]

Post by Jan Panteltje
96 dB is a lot, best of luck!

I've put some more information, including circuit diagrams at:
http://www.poppyrecords.co.uk/other/DistortionMeter/intermodmeter.htm

[Don]

Post by Liz Tuddenham
[...]

Post by Jan Panteltje
96 dB is a lot, best of luck!

http://www.poppyrecords.co.uk/other/DistortionMeter/intermodmeter.htm

Thank you for the link, Liz. Your website's always fun to visit - for
both its electronics and its music history. (One of my goals is to
improve my own website.)
My momentary muse happens to be fiddling with a crystal radio. Jan
suggested using a linear power supply for it. And now your excellent
intermodulation meter documentation serendipitously shows a suitable
schematic:

<Loading Image...>

I'll probably use some sort of linear regulator from the bone pile in
place of the transistors.

Danke,

[Liz Tuddenham]

Don <***@crcomp.net> wrote:

[...]

Post by Don
My momentary muse happens to be fiddling with a crystal radio. Jan
suggested using a linear power supply for it. And now your excellent
intermodulation meter documentation serendipitously shows a suitable
<Loading Image...>
I'll probably use some sort of linear regulator from the bone pile in
place of the transistors.

The voltage on the smoothing capacitors is perilously close to the 30v
maximum input voltage of most commonly-available voltage stabilisers,
that was why I used a descrete component pre-stabiliser instead of a
chip.

Most stabilisers have rejection factor of about 60dB, so interaction
between circuits on the same power supply rails isn't usually a
problem, but I was trying to measure interactions 90dB down, so each
section had its own stabiliser running off pre-stabilised lines that
were common to all of them. Two steps of 60dB are much easier to
achieve than one step of 120dB.

[Don]

Post by Liz Tuddenham
[...]

Post by Don
My momentary muse happens to be fiddling with a crystal radio. Jan
suggested using a linear power supply for it. And now your excellent
intermodulation meter documentation serendipitously shows a suitable
<http://www.poppyrecords.co.uk/other/DistortionMeter/images/PSU.gif>
I'll probably use some sort of linear regulator from the bone pile in
place of the transistors.

The voltage on the smoothing capacitors is perilously close to the 30v
maximum input voltage of most commonly-available voltage stabilisers,
that was why I used a descrete component pre-stabiliser instead of a
chip.
Most stabilisers have rejection factor of about 60dB, so interaction
between circuits on the same power supply rails isn't usually a
problem, but I was trying to measure interactions 90dB down, so each
section had its own stabiliser running off pre-stabilised lines that
were common to all of them. Two steps of 60dB are much easier to
achieve than one step of 120dB.

Thank you for sharing your thought process and your intermodmeter
documentation:

<http://www.poppyrecords.co.uk/other/DistortionMeter/intermodmeter.htm>

Empirical data indicates a zener-emitter-follower's rejection's superior
to a typical TL431:

<https://www.tnt-audio.com/clinica/regulators_noise3_e.html>

Integrated circuit complexity arguably adds noise, both in the literal
and the figurative, metaphorical sense.

Danke,

[Jan Panteltje]

On a sunny day (Sun, 25 Feb 2024 15:56:37 +0000) it happened

Post by Liz Tuddenham
[...]

Post by Jan Panteltje
96 dB is a lot, best of luck!

http://www.poppyrecords.co.uk/other/DistortionMeter/intermodmeter.htm

Nice
interesting method..
In the old days we did measure distortion in the Eurovision network audio channel by having the remote end (some other country)
send a 1 kHz tone, we then measured it and measured all the harmonics
2, 3, 4, 5 kHz etc with a selective volt meter.
And then did the math, distortion (harmonics) had to be below a specific level.
But nowhere near like -96 dB.. :-)

These days with PC sound card as generator, for example 'sox' in Linux will generate tones and sweeps either as output
or as file, ( have some 1 kHz files and sweep files over the audio range made with it)
All 16 bit 48 kHz sampling...
see:
https://www.audiosciencereview.com/forum/index.php?threads/howto-sox-audio-tool-as-a-signal-generator.4242/

There is also an audio spectrum analyzer program somewhere...
So any harmonics will be shown.
I really do not know where the audiophiles are as to noise level these days, or even if they are into bat frequencies.
Usually the environmental noise here when listening to audio is orders of magnitude higher ..

A good soundcard is a must, for example a Raspberry Pi with analog audio out (the old ones)
has a lot of RF as audio is generated by some sort of PWM, left unfiltered...
I have a very good soundcard I used for multi-channel multi-language audio processing in an old PC.
I have a small USB audio stick for in the Raspberry Pi too, it has audio out (line level) and audio in (mike),
but have not measured noise or output quality.
Sounds OK, was just a few dollars on ebay..
I did add a external lowpass on the analog audio output of the old Raspberry Pi to get rid of the RF.
OK for causal listening...

[Liz Tuddenham]

Post by Jan Panteltje
On a sunny day (Sun, 25 Feb 2024 15:56:37 +0000) it happened

Post by Liz Tuddenham
[...]

Post by Jan Panteltje
96 dB is a lot, best of luck!

http://www.poppyrecords.co.uk/other/DistortionMeter/intermodmeter.htm

Nice interesting method.. In the old days we did measure distortion in the
Eurovision network audio channel by having the remote end (some other
country) send a 1 kHz tone, we then measured it and measured all the
harmonics 2, 3, 4, 5 kHz etc with a selective volt meter. And then did the
math, distortion (harmonics) had to be below a specific level. But nowhere
near like -96 dB.. :-)

This equipment was intended for experimental work, rather than
transmission line-up, so it had to go a bit further than usual. It was
the result of 'mission drift' when I initially set out to disprove to a
valve fanatic's assertion that second harmonic distortion was
unimportant because it sounded "musical". (This is actually almost true
for a single pure tone, but not for complex music waveforms.)

I was looking for an easy way to measure intermodulation distortion when
I came across this idea in an article by Thomas Roddam in Wireless World
in the 1950s. I don't think he could have actually built one or he
would have discovered that his proposed iron-cored filter components
would have caused more distortion than they removed. I think the
original development of this method was done in the 1930s for the
American film industry, where a notch filter for THD would not have
worked because of the wow and flutter of the film sound track.

I do have a Marconi THD meter, which is basically a sharply tuned filter
that can be swept through the audio band using the superheterodyne
principle. As you say, it involves a lot of maths if you are interested
in each harmonic individually -- but it can be used 'the other way up'
to notch out the fundamental and measure all the other rubbish as a
single reading. Taking very low THD measurements by adjusting the notch
is an extremely slow and tedious business, compared with just operating
a couple of switches and reading a meter for the I/M method.

The great advantage of the THD method is that it gives one number which
the Sales Department can wave under the noses of the opposition, whereas
I/M measurements will differ according to the pairs of tones selected
(especially in amplifiers with iron-cored output transformers).

Post by Jan Panteltje
These days with PC sound card as generator, for example 'sox' in Linux
will generate tones and sweeps either as output or as file, ( have some 1
kHz files and sweep files over the audio range made with it) All 16 bit 48
https://www.audiosciencereview.com/forum/index.php?threads/howto-sox-audio
-tool-as-a-signal-generator.4242/
There is also an audio spectrum analyzer program somewhere... So any
harmonics will be shown. I really do not know where the audiophiles are as
to noise level these days, or even if they are into bat frequencies.
Usually the environmental noise here when listening to audio is orders of
magnitude higher ..

That is true, it always amuses me to see audiophiles comparing
recordings when their listening was done on a car radio. ...or
comparing amplifiers and loudspeakers using recordings that are almost
totally synthetic.

[Dave Platt]

Post by John Larkin
Nowadays one can do a DDS phase accumulator and a sine lookup, in an
FPGA or all software in a uP. Frequency and amplitude and even phases
and sweeps can be precisely programmed.
It does need a pretty good lowpass filter after the DAC to take out
the jaggies.

Yup. The faster you can run the DDS/DAC the better (all else being
equal, of course).

I put together something of this sort in an ICE40 FPGA, with a
serialized bank of DDS phase accumulators to generate sine and ramp
functions in one pipeline, a controllable-gain-per-input mixer in
another, and an FM-stereo multiplex encoder in a third. I didn't have
enough FPGA space to do a full 16-bit sinewave lookup, so that part of
the pipeline uses a coarser lookup with linear interpolation - as I
recall I got about 15.5 ENOB out of it. Most else is being done with
32-bit fixed point.

The output of the final mixer stage goes out serially to an audio DAC
running at about 350 kilosamples/second. It produces a very nice
FM-composite signal.

The output of the mixer is also added to a base increment, and the
result sent out serially (at the same 350 ksamples/second rate) to an
Analog Devices RF DDS.

The result is a very nice 10.7 MHz IF signal that I can use to analyze
and adjust FM tuner IF stages and discriminators and MPX decoders. By
mixing it with a 100 MHz carrier I can also test the tuner front end.

The very best tuner I've tried it with isn't actually intended as
such/ It's an old Racal-Dana modulation meter. This meter uses a
low-distortion pulse count discriminator, and has no IF filtering to
speak of (very wide bandwidth). I got 0.03% THD for a mono signal at
75 kHz deviation.

Quite a fun learning experiment (my first serious Verilog work).

[Liz Tuddenham]

Post by Jan Panteltje
On a sunny day (Fri, 23 Feb 2024 18:33:28 +0000) it happened

Post by Liz Tuddenham

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscilates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.
Interesting cases are HP, Walt Disney, and the first touch-tone
phones.

In an audio intermodulation meter I designed, the two tone sources had
to be as noise-free as possible. The problem was that any noise from
the gain-determining element of the high-frequency oscillator
intermodulated with the oscillator signal and appeared as if it were
amplitude variations. That set a limit to the maximum sensitivity of
the measurements.
After trying various configurations, I found that the quietest results
came from a Wien Bridge type of circuit where the gain was accurately
set to just over 3 and the op-amp was allowed to clip. The wavform and
frequency stability weren't brilliant, but that didn't matter.
As a noise-free way of controlling the amplitude, I generated
symmetrical low-noise power supply rails from op-amps driving
emitter-follower buffers, cntrolled by the voltage from a 'set level'
control on the front panel. Both oscillators were controlled by this
method, so their amplitudes tracked together.

In the late seventies I build an audio signal generator using
a 4046 voltage controlled oscillator feeding into 4040 binary counter
connected to an EPROM with sine lookup table feeding into an DAC.
You could make that as precise (as many bits) as you want.
8 bits was good enough for me :-)
It could frequency sweep audio circuits using a pot or some other source
on the 4046 VCO input.
Some lowpass against the switching noise..
Its easy....

What was the amplitude variation and noise after filtering? I needed at
least 96dB below signal and a few stages of filtering could actually
generate more noise than it removes.

[Bill Sloman]

Post by John Larkin
Among other parasitic functions, I'm the local LC oscillator designer.
A new board has a triggered 50 MHz oscillator, a Colpitts that uses a
SAV541 phemt. It has excess, erratic jitter that traces to a 6 GHz
parasitic oscillation. A couple of things fix that, including using a
Murata bead in the gate, thanks to a gift from Phil Hobbs.
I was thinking that there are two classes of LC oscillator, depending
on the amplitude limiting mechanism.
Many are AGC types (like the original HP audio oscillator) or a
circuit that self-biases off as amplitude goes up. That oscillates at
one frequency, where there's just enough gain to drive one mode.
But an oscillator may amplitude limit by some hard clipping, like a
diode clamp. When it's not clamping, there's full gain, basically
negative resistance, to excite any parasitic resonators. The mixed
oscillation mode is sometimes called squegging.

"Squegging" is an ill-defined term, Essentially it is a chaotic
oscillation, which can repeat exactly but frequently doesn't.

<snip>

Hard clipping always generates higher harmonics - mostly all of them up
to a limit set the period in which each cycle is clipped. Those
harmonics can excite other resonances.

I've explored the idea of using a four quadrant multiplier -
specifically the AD734 (despite it's ridiculous price) to control the
amplitude of a Wein Bridge. LTspice suggests that it would work rather
well. The AD734 generates harmonics at around the -70dB level but when
used as a gain correction mechanism the correction signal should be 60dB
below the output, so the consequent harmonic level in the output should
be better than 130dB below the fundamental.

I've also explored the idea of using a second - in quadrature - feedback
loop to correct the frequency of the oscillation. That worked quite well
in LTSpice, after some consultation with my friend in Scotland. Getting
the quadrature signal required a somewhat messy phase shift network

You should be able to build an instant-start version of the circuit, but
the jitter is never going to be as low as you can get with a faster
oscillator running non-stop.