In building a music synthesizer with the LM3900, I've been finding increasingly better approaches that offer good precision out of the LM3900 Norton op amp. This is notable, because the LM3900 is certainly not a precision op amp. All the same, investigating it's deeper secrets is providential to getting good performance instead of mediocre or average.
A key area for the synthesizer is a VCO. For the LM3900, the AN-72 VCO is rather a classic design, albeit not neccessarily as-is for music. But as an element for a PLL, or for a narrower range of frequencies, the LM3900 enabled a nice alternative to a much more expensive LM566 or LM567. This was the very first VCO I ever attempted, using a LM3900 I obtained from a Radio Shack store in Portland, Oregon, in the mid-1970s. This was relatively soon after the release of the LM3900 by National Semiconductor in the Fall of 1972.
Also, I'm specifically aiming to use this op amp to obtain usable results, not record breakers for dynamic range, etc. For sure much higher performance voltage op amps exist, but I'm specifically working to build a synthesizer using with a unique sound by applying industrial control oriented parts from the 1970s like the LM3900. Ultimately a 5 octave range, or 32:1 frequency range, would be adequate. This represents 60 notes on the equal-tempered scale. Plenty.
Of course when building the AN-72 VCO for a music application, one discovers some problems right away. A first one is with dynamic range; another is with triangle wave symmetry for lower control voltgages; an another is that the +Vbe present at the inputs to the LM3900 requires a control voltage offset from ground by a like amount in order to obtain a 0 Hz output.
But the topology of the AN-72 circuit is quite interesting, a current switching one, where a 2I current is ratioed against a 1I current by using the current mirror at the +input to subtract current from the -input. Using the unique current differencing capability allows the LM3900 to perform differential integration. In this case, the current mirror performs 1I - 2I = -1I, which provides the current reversal for positive integration that allows the up slope integration of the triangle wave generation core. The down slope integration is done soley with 1I, as during that half of the period, the current mirror at the +input is shut down.
This new circuit keeps the current differencing topology, but uses input currents directly instead of voltage-to-current generation directly using resistors at the LM3900 inputs, per the AN-72 VCO. This technique of direct current inputs has already proven possible with an envelope generator that had very wide dynamic range using both the +input and the -input.
The technique used here for current generation was to use two Howland Current generators, each as a Voltage Controlled Current Source (VCCS), one for 1I, the other for 2I. A LM324A voltage op amps was used for this, with the current setting resistors set according to the 1:2 need. Use of the AN-72 topology, where the +input current mirror is shut off for half of the triangle wave generation cycle, but for the other half-cycle sources 2I to be subtracted from the 1I. This mode is providential over the -I/+I differential integrator technique for the envelope generator, which used a CMOS switch. This is because the CMOS switch has charge injection, and with integration capacitors sized for audio frequency, the waveform fidelity is terrible.
The Howland circuit did work rather well, but one aspect is that the LM32A op amps need ±15V power supplies. The Howland circuit is also only really being "half-utilized," and is capable of more than is needed here: it can perform bidirectional current generation. Instead of the ratioed unidirectional current generation needed for the LM3900 VCO. So, there's a little bit of complexity that is undesired, including use of a -15V power supply voltage. A future form of this circuit will be exploring using the LM3900 for 2I/1I current generation, and only a +15V power supply voltage.
This variant of the AN-72 VCO used a Schmitt Trigger with rather wide trip points, which causes the triangle wave generated to have wide swing > 10V. As a result, the upper frequency range of the VCO will be much more limited than in the case where the Schmitt Trigger limits the integrator to a ΔV ≤ 2.5V. As well, the component values for the Schmitt Trigger applied smaller currents to the LM3900, which results in slower operation, as well as additional sensitivity to stray signals like AC hum.
Given the centrality of the Schmitt Trigger to this VCO, additional research has been conducted for improved performance. First a much faster Schmitt Trigger was developed, using larger currents and a smaller UTP - LTP = ΔV ≤ 2.5V. Secondly, a more precise Schmitt Trigger was subsequently developed, where the UTP and LTP have precision voltage values independent of temperature and power supply voltage, and where the ΔV ≡ 2.5V precisely. This second variant will feature in a new LM3900 VCO design that is much more suitable for music synthesis.
Some scope fotos follow.
Comments
Post a Comment