Not exactly HiFi audio as this is specifically designed for testing the IMD performance of SSB radio transmitters, but it contains another example of a low distortion generator using an LDR for amplitude stabilization for interest.
I got tired of adjusting and passively summing the outputs of a pair of bench function generators every time I need to measure or test a linear amplifier and decided that a dedicated generator would be a lot more convenient. A typical linear power amplifier will struggle to generate 3rd order distortion products much better than 40dB below reference carrier and the 5th order products maybe 10dB better than that, so the linearity requirements for the test oscillators are not particularly onerous.
So, the bench generators have been OK for now, however I am now building a series of low-level SSB exciters and the low-level circuitry is easily 20dB better, so a two-tone generator better than the typical designs published and built by HAMs is required here. Most of these circuits have THD levels measurable in the tenths of a % and are only really useful for visually inspecting the basic SSB envelope linearity on an oscilloscope and completely useless for measuring IMD on a spectrum analyser.
Not wanting to do what everyone else has done before, I thought I'd try something different. I used a multiple-feedback (MFB) bandpass filter for the frequency-determining block rather than the typical Wien bridge. I have pretensions of calling this design a lab-grade instrument, so a requirement was that the spot frequencies and peak amplitudes need to be adjustable for calibration. A neat property of the MFB bandpass filter is that the centre frequency can be adjusted independently of gain by varying a single resistor.
I initially breadboarded the oscillator design with ordinary AD711 DIP op-amps and got in the order of .001% THD running at a couple of kHz. I figured that was more than good enough for the application and quit experimenting. This was with a designed-in gain of unity for both the MFB bandpass filter and the inverting loop amplifier.
However, once I got the completed PCB design loaded and running, I was disappointed at the 700 Hz oscillator performance. The 1900 Hz oscillator did well at approx. 0.0005% THD, but the 700 Hz oscillator was much higher at around .004%. In all honesty, 0.01% would be more than good enough for the intended application, but this is not a result that I could leave alone.
I initially theorised that the higher distortion must be due to a higher level of rectifier ripple modulating the amplitude control loop due to the lower operating frequency, but that was ruled out after a temporary fudge of some component values to reduce the ripple to significantly less than that of the 1900 Hz oscillator - the distortion performance remained completely unchanged.
The worse linearity of the 700 Hz oscillator as entirely due to the NSL-32SR2 LDR opto-coupler. But why ~8x worse for one oscillator over the other? There are only two possibilities:
1) Huge performance variance between parts.
2) LDR linearity is strongly frequency dependant.
I would find 1) to be less surprising than 2) but cannot rule either out until I have done some further experimentation. Any others out there have experience here?
Anyhow, all of this LDR distortion stuff is now moot as far as the actual design is concerned as I have practically eliminated LDR distortion with the changing of a bunch of component values. I reconfigured the MFB bandpass filters to have a gain of 0.2 rather than unity and the inverting loop amplifiers for a gain of 5 instead of unity. This a just a gain distribution change to drop the ac voltage developed across the LDR by factor of 5 whilst retaining a complete loop gain of unity as required for oscillation. Oscillator outputs are taken directly from the MFB bandpass filters as this is beneficial for distortion, but as the amplitude here dropped by 5 times, I had to increase the gain of the summing amplifier accordingly.
Several hours of component value tweaking later and I now have both tone oscillators challenging the distortion performance of by QA403 Audio Analyser. I think I am done tweaking this design now. The schematics that follow show the final component values. I have a few through-hole resistors standing over 1206 footprints, but that is only a temporary measure to substitute for my current lack of MELF stock of the required values. The LM380 and speaker is for conveniently singing directly into the transmitter's microphone when all you are doing is a basic operational health check of the transmitter by looking at the resultant modulation envelope on an oscilloscope.
I got tired of adjusting and passively summing the outputs of a pair of bench function generators every time I need to measure or test a linear amplifier and decided that a dedicated generator would be a lot more convenient. A typical linear power amplifier will struggle to generate 3rd order distortion products much better than 40dB below reference carrier and the 5th order products maybe 10dB better than that, so the linearity requirements for the test oscillators are not particularly onerous.
So, the bench generators have been OK for now, however I am now building a series of low-level SSB exciters and the low-level circuitry is easily 20dB better, so a two-tone generator better than the typical designs published and built by HAMs is required here. Most of these circuits have THD levels measurable in the tenths of a % and are only really useful for visually inspecting the basic SSB envelope linearity on an oscilloscope and completely useless for measuring IMD on a spectrum analyser.
Not wanting to do what everyone else has done before, I thought I'd try something different. I used a multiple-feedback (MFB) bandpass filter for the frequency-determining block rather than the typical Wien bridge. I have pretensions of calling this design a lab-grade instrument, so a requirement was that the spot frequencies and peak amplitudes need to be adjustable for calibration. A neat property of the MFB bandpass filter is that the centre frequency can be adjusted independently of gain by varying a single resistor.
I initially breadboarded the oscillator design with ordinary AD711 DIP op-amps and got in the order of .001% THD running at a couple of kHz. I figured that was more than good enough for the application and quit experimenting. This was with a designed-in gain of unity for both the MFB bandpass filter and the inverting loop amplifier.
However, once I got the completed PCB design loaded and running, I was disappointed at the 700 Hz oscillator performance. The 1900 Hz oscillator did well at approx. 0.0005% THD, but the 700 Hz oscillator was much higher at around .004%. In all honesty, 0.01% would be more than good enough for the intended application, but this is not a result that I could leave alone.
I initially theorised that the higher distortion must be due to a higher level of rectifier ripple modulating the amplitude control loop due to the lower operating frequency, but that was ruled out after a temporary fudge of some component values to reduce the ripple to significantly less than that of the 1900 Hz oscillator - the distortion performance remained completely unchanged.
The worse linearity of the 700 Hz oscillator as entirely due to the NSL-32SR2 LDR opto-coupler. But why ~8x worse for one oscillator over the other? There are only two possibilities:
1) Huge performance variance between parts.
2) LDR linearity is strongly frequency dependant.
I would find 1) to be less surprising than 2) but cannot rule either out until I have done some further experimentation. Any others out there have experience here?
Anyhow, all of this LDR distortion stuff is now moot as far as the actual design is concerned as I have practically eliminated LDR distortion with the changing of a bunch of component values. I reconfigured the MFB bandpass filters to have a gain of 0.2 rather than unity and the inverting loop amplifiers for a gain of 5 instead of unity. This a just a gain distribution change to drop the ac voltage developed across the LDR by factor of 5 whilst retaining a complete loop gain of unity as required for oscillation. Oscillator outputs are taken directly from the MFB bandpass filters as this is beneficial for distortion, but as the amplitude here dropped by 5 times, I had to increase the gain of the summing amplifier accordingly.
Several hours of component value tweaking later and I now have both tone oscillators challenging the distortion performance of by QA403 Audio Analyser. I think I am done tweaking this design now. The schematics that follow show the final component values. I have a few through-hole resistors standing over 1206 footprints, but that is only a temporary measure to substitute for my current lack of MELF stock of the required values. The LM380 and speaker is for conveniently singing directly into the transmitter's microphone when all you are doing is a basic operational health check of the transmitter by looking at the resultant modulation envelope on an oscilloscope.
Last edited:
