Triangle / Sawtooth VCO with voltage controlled

continously variable symmetry

Achim Gratz
March 26, 2006
http://homepages.compuserve.de/Stromeko/diy/SawWM.pdf

mailto:[email protected]

Draft Copy!
Permission granted to link to the URL given above.
Do not make publicly available copies (aka mirroring).

A discussion of how to get continously variable symmetry (ratio of the up and the down part to
the period of the waveform) from a sawtooth or triangle VCO.

1 Objective
The circuit should be able to produce a waveform that is continously variable via a control voltage
(CV) from sawtooth through triangle through to inverted sawtooth while maintaining the frequency
set independently by another CV. Without loss of generality we assume that the CV VW controls

V V V

t t t

0 T 0 T 0 T
Saw wave (up): Triangle wave: Inverted saw wave (down):

1
VW = VW,min VW = VW,max VW = VW,max
2
1
w= 0 w= w= 1
2
v = −1 v= 0 v = +1

Figure 1: Waveforms at different values of the control voltage and modulation variables

the ratio of the downward slope time versus the period so that at zero control voltage the waveform
becomes a sawtooth, a triangle wave is produced at VW = 21 VW,max and an inverted sawtooth is
produced at maximum VW . Instead of dealing directly with the CV we introduce a modulation

1
variable w that is normalized to the maximum CV.
VW
w= | VW ≥ 0 (1)
VW,max
T = tu + td = (1 − w)T + wT | w ∈ (0, 1) (2)

Alternatively we may consider a triangle wave at zero modulation and modulate towards a saw
or inverted saw with positive and negative modulation, respectively. The modulation variable is
bipolar in this case.

2VW − (VW,max + VW,min )

v= (3)
VW,max − VW,min
1+v 1−v
T = tu + td = T+ T | v ∈ (−1, 1) (4)
2 2

2 Dual Slope Integrator VCO

A sawtooth integrator VCO charges a capacitor with constant current ISaw until an upper threshold
voltage U^ is reached. The capacitor is then discharged as quickly as possible to the lower voltage Ǔ.
For a triangle VCO twice the current is used into the integrator capacitance for the same oscillation
frequency and upon reaching the upper threshold the direction of current is reversed to discharge
the capacitance to the lower threshold. We are going to use a dual slope integrator with different
currents for the charge and discharge phase.
U^ − Ǔ
T= (5)
CISaw
U^ − Ǔ
tu = (6)
CIu
U^ − Ǔ
td = (7)
CId
T ISaw = tu Iu = td Id (8)

in particular

2ISaw = ITriangle (9)

and with (2), (3) and (9)

ISaw = (1 − w)Iu = wId (10)

ITriangle = (1 + v)Iu = (1 − v)Id (11)

which leads to
ISaw ITriangle
Iu = = (12)
1−w 1+v
ISaw ITriangle
Id = = (13)
w 1−v
Iu w 1−v
= = (14)
Id 1−w 1+v
(1 − w)Iu (1 + v)Iu
1= = (15)
wId (1 − v)Id

further to
ISaw
IΣ = Iu + Id = (16)
w(1 − w)
2ITriangle
IΣ = Iu + Id = (17)
1 − v2

2
 and finally to

Iu = wIΣ = (1 − v)IΣ (18)

Id = (1 − w)IΣ = (1 + v)IΣ (19)

We observe that the charge and discharge currents are ratiometric to each other in proportion to
the applied modulation, their sum is always larger than four times the charge current of a simple
sawtooth generator while the individual currents can never become smaller than the charge current
of the sawtooth generator. Errors in the up and down charge currents will change the symmetry
and the frequency of the oscillator. It is possible to trade the frequency error in for an amplitude
error by means of an alternative control circuit. In both cases the error can be converted to a signal
that could be used in a feedback loop to correct it.
It also becomes clear that the modulation range has to be restricted as otherwise the sum current
(through one of the individual currents) would become infinite. Independently of how the charge
and discharge currents are produced, it proves quite difficult to keep the frequency constant through
the modulation. However if additionally triangle and/or sawtooth waves are produced from the
same CV it might be more interesting musically to let the phase of the oscillators drift with respect
to each other instead of locking them in DCO style.

3 Dual Sawtooth Waveshaper

The waveshaper uses an arrangement of zero based1 up and down sawtooth of the same frequency
and phase. These are amplified by an appropriate amount and the output waveform is produced
as the minimum of the two amplified sawtooths. Hence

t mod T
VSaw,u = au (20)
 T 
t mod T
VSaw,d = ad 1 − (21)
T

are up and down sawtooth waveforms with amplitudes au and ad , respectively. These two quan-
tities can also be regarded as gain factors for sawtooths with their amplitude normalized to one.
The output waveform is obtained as

VOut = min(VSaw,u , VSaw,d ) (22)

The switch from the up to the down sawtooth happens when their momentaneous amplitude is
equal, which leads to

tud mod T ad
= (23)
T au + ad
au ad
Vout (tud ) = (24)
au + ad

In order for the output amplitude to stay constant at one it is necessary for the two gain factors
to satisfy

1 1
1= + (25)
au ad

Thus the up and down times of the combined waves are

1 1
T= T+ T (26)
au ad

1 It is possible to keep them symmetric around zero, but then one needs to offset by 2aVpeak when a gain of a is applied.

3
 With (2) and (3) we get

1 2
au = = (27)
1−w 1+v
1 2
ad = = (28)
w 1−v
which we find to correspond to (12) and (13) normalized to ISaw . This is welcome news as it means
the two methods can use almost the same circuitry and all the derivations in 2 can be used by
simply normalizing them to ISaw . Note that by construction any errors in one or both of the gains
alter both the amplitude and the symmetry of the output wave, but not it’s frequency.

4 Implementation
Don Tillmans’s brilliant idea2 saves the multiplier otherwise necessary to satisfy (25). In the
context of the waveshaper this keeps the amplitude constant throughout the modulation. Don
Tillman observed that the identity
1 1
1= x
+ (29)
1+e 1 + e−x

holds unconditionally and leads to

au = 1 + ex (30)
ad = 1 + e−x (31)

This has an almost trivial implementation as a bipolar transistor circuit. However, as we have seen
above this circuit does not only work in the context of the waveshaper – we could also use it to
produce the two currents for a dual slope integrator VCO.
As proposed originally, however, there are no provisions for linear response of the symmetry
modulation, more specifically the modulation gets slower at the end of the modulation range. To
make the modulation linear with the CV we use (27) and (28) to arrive at

w 1−v
x = ln = ln (32)
1−w 1+v
which again has a straight-forward circuit implementation. Putting the two together we find that
we’re implementing a special configuration of a Log-Antilog-Multiplier. We cannot use an analog
multiplier IC directly, however, as for instance the RC4200 would only implement (15), which is
underdetermined as we can fix only two of the four variables. This means we’d need to use yet
another multiplier to determine the third variable, for instance to calculate (12).
Instead, we can use a discrete log-ratio amplifier to obtain
1+v
xd = −VT ln (33)
1−v
1−v
xu = −VT ln (34)
1+v

simultaneously by using just one additional opamp, then proceed to de-logarithmize this

−xd 1+v
yd = Uref exp( ) = Uref (35)
VT 1−v
−xu 1−v
yu = Uref exp( ) = Uref (36)
VT 1+v

2 http://www.till.com/articles/VariableSaw/index.html

4
 Uref
and finally add one in the form of Uref to obtain

2Uref
Ud = (37)
1−v
2Uref
Uu = (38)
1+v
which compares favorably with (12), (13) and (27), (28). A rough sketch showing a circuit along
those lines is appended. None of the necessary compensation elements for the opamps are shown.
However there is no overall temperature dependance to first order if all transistors are kept at
equal temperature. Note that by replacing the difference amps of the log-ratio amplifier by two
two-quadrant multipliers with their gain input tied together, the response curve of the symmetry
modulation can be continously varied from sub- to superlinear.

5
 Uref
1
-VT ln( 1-v / 1+v )
1-v
v -v
2Uref / 1+v

6
2Uref / 1-v

1+v
-VT ln( 1+v / 1-v )