You can get greater accuracy and less drift by deriving the sine
wave from a digital source. Because square waves comprise a
fundamental at the square-wave frequency plus an infinite number of
odd harmonics, you can obtain the desired fundamental sinusoid by
removing the harmonics with a lowpass filter. Switched-capacitor
filters suit this application **(Figure 1)**. IC3 is an
8th-order, lowpass Butterworth type.

*Figure 1. Filtering the harmonics from a square wave produces
a sinusoidal output whose stability and flexibility derive from
digital circuitry.*

The sine-wave generator starts with an 8MHz signal and divides it
by eight to obtain 1MHz at C1. (IC1's 2MHz and 500kHz outputs can
serve as alternate drive signals.) Q1 level-shifts the 1MHz pulses
so they can drive the bipolar circuitry necessary for producing a
bipolar output. (For unipolar outputs, you can operate the circuit
on a single supply voltage by biasing the IC3 ground terminal to
mid-rail and adding a decoupling capacitor.) Synchronous counter IC2
divides 1MHz by 256 to give the desired output frequency (3906Hz),
and IC3 filters the harmonic frequencies.

The filter's clock is taken from the first divide-by-2 tap of
IC2, to assure a 50% duty cycle. IC2 further divides this signal by
128 to assure that the filter's input signal (1MHz/256) falls within
the flat portion of the filter response. Fifty-percent duty cycles
on the IC2 outputs assure a symmetrical sine-wave output. The
filter's major pole, or corner frequency, is fixed with respect to
the clock and forms an internal clock-to-corner ratio of 100:1.
Filter attenuation lowers the third-harmonic amplitude to -80dB.

Because the filter's input and clock frequencies have a fixed
ratio of 1:128, switching or sweeping the frequency applied at C1
has a proportional effect on the sine-wave generator output.
Switching this frequency from 2MHz to 500kHz, for example, switches
the output frequency from 7812Hz to 1953Hz. Output amplitude is not
affected because this band is well below the smoothing filter's
25kHz corner frequency. Alias frequencies are not a problem, because
the frequencies that represent a potential cause of aliasing in this
circuit-the odd-numbered harmonics that exceed half the clock
rate-have insignificant amplitudes.

*This article is from the Maxim Website. A related idea
appeared in the 7/25/94 issue of Electronic Design.*