Improvements on VCO

Magnus Danielson magnus at analogue.org
Sat May 16 21:58:02 CEST 1998


Hi!

I have been thinking of how to improve the ASM-1 VCO.

There is several issues and I thought that I would bounce some ideas
with you guys....

The VCO is not very sensitive to DC supply voltage but it is sensitive
to AC supply current. This works the following way:

The VCO has two points where supply voltage will affect
frequency/tracking/jitter. They are:

1) The reference current generator that is formed by the 150 kOhm
   resistor from +15V to one of the matched transistors. It is
   intended to generate an reference current of 100 uA.

2) The voltage reference divider formed by a 20k/10k resistor divider
   from +15V to ground. The output of this divider is +5V which is
   used by the LM311 to compare the ramp voltage so that the LM311 can
   trigger the JFET to reset the capacitor.

As it happends, these two will compensate each other for DC voltage
errors of 15 V. It will be tuned for 14V and 16V just as well.
However, for any AC voltage also existent between ground and +15V the
VCO will feature two different dependecies that ougth to get fixed.

The AC components in the current generator will make the current
differ and therefore will the accumulation in the capacitor vary
accordingly. This will directly be overlaied onto the waveform itself!
The errors in waveform will also introduce jitter in the resetpoint.

The AC component in the comparator will also introduce jitter in the
resetpoint, the trouble is that I don't think that the jitter
introduced by either if the jitter sources will fully compensate each
other... I think that we can see intermodulation here!

One cannot fix one of these two points without fixing the other,
eg. both should derviate the same voltage in order not to loose the DC
tracking property. The nice thing here is that neither of them requre
any large current, thus they may both be connected directly into a
voltage reference. One thougth was to have a REF-02 5V reference to
supply both points with reference voltage. Using a REF-02 (which can
supply more than 21 mA) we will need to replace the 150 kOhm resistor
with a 49.9 kOhm resistor and the 20k/10k resistor with a 6.8 kOhm
resistor.

The 6.8 kOhm resistor would go from the REF-02 output to the LM 311
input and the input capacitor for sync. The trick here is that the
timeconstant has not been changed and the 6.8 kOhm resistor will have
the same output resistans and output voltage as the old divider
cursuit had and thus not change the impedance and sourced voltage seen
by the capacitor and the LM311.

I have noticed that the LM311 and JFET will cause a ringing when they
are turned off. This ringing is very high frequent and will travel
over the powerlines. One clue is that the fall-time on the LM311 is in
the neightborhood of 25 ns for a 15V jump, this currentspike is
certainly sharp enougth to cause a ringing in nearby supplylines.
The ringing is certainly there in the output waveform. I think that
some highspeed bypass capacitors needs to be applied. These spikes can
cause intermodulation between the VCOs. I have a picture of the reset
curves showing this!

One could ease up the adjustment procedure quite easy. Today one has
to fiddle about on the scale and offset pots in a repetive pattern and
apply some (unknownly exact) external voltage to jump between known
reference frequencies. While this is a bit tedious there is luckilly
simlpe means to fix it and make the adjustment procedure both simpler
and more exact.

One error that the normal adjustment methods of the ASM-1 does not
directly address is the frequency error that occur due to imprecission
in resistor and capacitor values. This will cause the unmodulated
frequency of the oscillator to be scaled diffrently. The unmodulated
frequency is the frequency where the output of the CV summing amp is 0V.
This rest frequency is dominated by the integrator capacitance and
reference current and those the resistance of the reference resistor.
This frequency is:

      3         3
f  = --- = ---------- = 909.0909 Hz
 0   R*C   150k * 22n

The output frequency using CV modulation will be:

           3 * Vlin      -0.6828*Vsum
f = (f0 - ----------) * e
          Vcc*C*Rlin

Where Vsum is the summation of the modulating CV wich is scaled and
offset.

Vsum is slightly scaled to make the exponential term act as a power of
2 function (we double frequency every octave, that is every 1 Volt).
As it happends will the CV offset correction also be scaled, and when
we correct the scale will the offset be slightly moved and when we
adjust the CV offset will the other tone for the scale be missplaced.
This cause an repetetive adjustment round until both the offset and
scale corrections have converged. Another problem is that the
potentionmeters have a too large adjustment range than migth be
necessary (thus causing a need for large initail adjustments to come
close to correct operations).

Fortunatly is there several ways to simplify this adjusment
procedure. The offset is scaled by the scale potentiometer, which is
not optimal, we would like to change the linear offset of the
oscillator in a diffrent way. One way would be to move the CV offset
trimpot over to the positive input of the summing amplifier, this way
would the CV offset be unscaled, the bad thing about this is that you
will loose a pure CV adjustment method. One could add a trimpot at the
reference current, this would improve things.

With the reference current trimpot one would adjust this way:

* Adjust CV so that the summing op-amp outputs 0V
* Adjust the reference current trimpot so that the oscillator hit A5 (880 Hz).
* Apply 1V to CV input and adjust the scale trimpot for A6 (1760 Hz)

The last step can be done with higher voltage steps to get better
scale precission.

The range of the trimpots could possibly the reduced so that the
trimpot will only operate over the range where one can expect
variations. This way can the pots positional precision be better used.

One things that I have been thinking of is that the tempco resistor
will not be very active when the voltage applied to the 56k/1k
division cursuit is 0. This may be due a failure in analysis (I have
just thinked briefly about it and not done real analysis) or an actual
misstake.

The linear FM inputs is as you can see negative, so that a positive
voltage will cause an negative frequency change. The depth of the
linear FM will also be scaled with the CV. The scaled FM will work
such that a linear FM voltage that caused a 10 Hz offset at 0V CV in
will cause an 20 Hz offset at 1V CV in. One can easilly create an
unscaled linear FM input where an summing op-amp would apply a voltage
onto a resistor that connects to the current input of the integrator.
An positive linear FM input can be created by using the positive input
of the linear FM input summing amp.

Another thing that could possibly be a slight trouble is that Sawtooth
output will have a slight DC offset. The DC offset is also dependent
on the negative powerrail and dirt on the negative powerrail will make
it to the Sawtooth output with an amplification of -200/560 or -8.94
dB damping. Removing the 560k resistor and tying the positive input of
the Sawtooth output summer to the REF-02 +5V output will fix this.
This fix migth cause a little trouble thougth, but this is due to the
ASM-1 PCB and not pure electrical/component wise.

When I measured up one of my ASM-1 VCO's I hit into two limits in
frequency, both a upper and a lower:

* Upper frequency stopped at 102 kHz after going off-track severly.

Up to a certain point will the VCO track just as it is supposed to and
then abruptly (somewhere after 81 kHz) will it not follow along and
you migth force it somewhat more but not longer than 102 kHz. I think
that this is due to the fact that the MAT-02 transistor is reaching
saturation and this is the limiting factor. This happends very
abruptly and it seems like the tracking nearby the abrupt area is not
greatly affected. To go higher would a redimensioning of the
oscillator core be necessary.

The limiting max current that the MAT-02 could deliver in this setup
seems to be

I      = f    * V    * C = 102k * 5 * 2.2n = 11.22 mA
 C2max    max    top

which is 112.2 times the reference current of 100 uA and 102k/880 =
115.9 times in frequency from the unmodulated frequency (for which the
reference and output currents are equal).

A look into the MAT-02 datasheet say 20 mA as Absolute maximum, so
this seems quite correct. All signs say that one should not use the
MAT-02 above 10 mA, so the dimensioning formula should be:

10 mA = f    * V    * C
         max    top

Since we settled for Vtop = 5V this turns into

      10m       2m
C = -------- = ----
    5 * f      f
         max    max

* Lower frequency stopped at 88 mHz with no tracking problem prior to
  that.

This showed out to be caused by the CV summing op-amp could not
deliver more than +14.2 V. To get lower would a redimensioning of the
oscillator core be necessary.

Rescaling the core need to take these limits into account. By chaning
the reference current resistor one can also adjust the center
frequency according to

     3          3
f  = -- => R = ---
 0   RC        f C
                0

which when combined with the dimensioning formula for C becomes

    f           f
     max   3     max
R = ---- * -- = ---- * 1.5 k
     f     2m    f
      0           0

This formula will make it simple to control the range that is
available above the rest frequency.

I migth add that when you go higher in frequency the cursuit will go
of in frequency tracking due to the fixed reset time.

Another high-frequency error is that the reset will not reach the
bottom region. I think that this is caused by the combination of the
reset resistor of 680 Ohm and the pulling current. Reducing this
resistor will reduce the floor level effect at high currents.
The floor-raise is almost up to 1V on the 5Vtt sawtooth.
I have this floor level on picture too!

These thougths on improvements can be neatly be combined with further
ideas on sync-improvements, further waveshapers etc. Without any
conflict (depending on your sync change that is).

I have been thinking of maybe have a PLL function just like the Moog
921's have. This would feature a 4046 or maybe just a XOR gate, some
RC filter and a buffer. I don't think one have to get advanced on
it...

Cheers,
Magnus



More information about the Synth-diy mailing list