When a curious person searches the Internet for the circuit diagrams of (electronic) mixers, there is a certain complexity of what he or she will find. Just for people who might not know, the type of mixer I’m referring to is a component which does not add two signals together – which is what the naming might seem to suggest – but rather, which multiplies two signals. In certain cases the mixer will produce output, that contains an additive component as well as a multiplied component. But it’s the multiplied component circuit designers are interested in, because that can be used:
- In order to produce ‘mixed frequencies’, between two input frequencies, such as between a local oscillator and a Radio Frequency, resulting in an Intermediate Frequency,
- In order to act as a phase discriminator, the output of which will be maximally positive or negative, when two input signals are in-phase, but the output-voltage of which will be some neutral voltage, when the input waves are 90⁰ out-of-phase with each other. In this latter case, two reasonably constant input amplitudes are assumed.
What search results will often show, is somewhat complex mixers, that require either one or two balanced inputs – meaning inputs conditioned such, that they each appear differentially between two input electrodes – and which have as advantage for being designed that way, low distortion of the wave-form(s) supplied differentially in this way.
But sometimes, low distortion is not required. For example, in the case of a PLL – a “Phase-Locked Loop” – It’s assumed that the feedback voltage changes the frequency of a VCO – a “Voltage-Controlled Oscillator” – but with the intended result that two outputs lock in some phase-position, so that the two frequencies that are inputs to the phase-discriminator will be exactly the same frequency. This latter need often arises in the design of ICs. This latter application does not require that the phase-discriminator be particularly linear, nor that its output-voltages, that become feedback voltages, be in any range other than the range which the VCO requires as input.
And so the question can arise, what the simplest way might be to design a mixer, with the added detail that both inputs are unbalanced inputs – i.e., that each input appears at one terminal, and not in an opposing way, at two terminals – and for the sake of argument, our IC might be limited to using enhancement-mode, N-channel MOSFETs as the main active component. And this would be my solution:
The concept is very simple. If Vin1 and Vin2 are at 180⁰, then M1 and M2 don’t conduct simultaneously. Therefore, R1 and Vcc (the supply voltage) achieve maximally positive average output-voltage. If Vin1 and Vin2 are at 0⁰ phase-position, the two transistors will become conductive in a way that coincides. Therefore, this is actually a Coincidence Detector. And the average output-voltage will be maximally negative in that case. And, if Vin1 and Vin2 are at a 90⁰ phase-position, then the average output-voltage will be somewhere between the two values mentioned before.
I suppose it should be mentioned that, if the circuit designer knows ahead of time that one of the two inputs has a much higher amplitude than the other, or a more predictable amplitude, then this usually stronger input should be fed to Vin1.
As part of a feedback loop, the output needs to be followed by a low-pass filter, that emulates an integrator over the time-constant which is the fastest, with which that feedback loop is supposed to be able to react to a change in one of the frequencies. The simplest low-pass filter consists of a resistor followed by a capacitor… (:1)
And so, when looking for a way to implement a phase-discriminator, the curious person needs to choose which of the following has greater priority:
- The simplest circuit-design, or
- The lowest amount of distortion.
The circuit above will certainly give the highest amount of distortion.
(Updated 7/9/2019, 16h55 … )
(As of 7/7/2019: )
Especially since I’m focusing on the context of ill-suited MOSFETs, working at a supply voltage of only 3V, the fact becomes apparent that there will barely be enough of a voltage range, between Vout-Average and the threshold voltage of the MOSFETs, to switch a MOSFET between being fully conductive and being fully non-conductive. Simultaneously, the type of VCO I’m considering – an “Astable Multivibrator” – will tend not to draw any long-term, DC current from the source of their control voltage, instead only subtracting and then adding short-term pulses of current, if the use of MOSFETs is substituted for the use of the bipolar transistors, which the linked article suggests. Other VCOs tend to draw considerable current from the source of their control-voltage, requiring that R1, as it stands, have a lower resistance value.
This set of problems can be mitigated, by connecting a capacitor of about 100pF directly between Vout above and ground, which will act as a low-pass filter together with R1 (implying a corner-frequency of 16kHz, to define the speed of the feedback loop). This will achieve maximum frequency-range from the VCO.
(Update 7/9/2019, 16h55 : )
The best policies for biasing M1 and M2 are:
- Wherever the waveform will have a ‘high’ amplitude, such as 0.5V peak, the MOSFET should be biased just-off,
- If M2 is to receive a ‘weak’ signal, such as an RF signal, then it should be biased just-on.
The best bias-voltage may be determined by a kind of ‘analog computer’, that simulates the circuit to be biased (see below :2).
(As of 7/7/2019 … )
The situation becomes more interesting, if the available frequency-range of the VCO is to be constrained. In that case the problem with the use of any VCO becomes, that the supply voltage cannot be guaranteed to be perfectly accurate, nor as accurate as the assigned frequency-range is supposed to become. And this problem can only be solved, through the use of an active circuit that’s to act as a controlled voltage-reference, such as a 2.4V Zener Diode (Discrete Component Alert!). The Astable Multivibrator above could be connected to this voltage-reference via 1.5kΩ, but to the output of the mixer as well, the latter via a resistor that can be made arbitrarily high, such as 100kΩ…
When working with Vcontrol = 2.4V, VT0=1.8V:
F = 1 / (2 * ln|3| * R * C)
~= 1 / (2.1972 * R * C)
The thought has occurred to me that there could be a better way, to constrain the frequencies of such a VCO. The premise could be that an AM/FM radio uses the same IF strip, and the same final IF centre-frequency (say 450kHz), for FM reception that it does for AM. This means that it should have a demodulator available, that produces a result of ‘zero’, when its received frequency is exactly 450kHz.
Well, when such a radio is operating in AM mode, the carrier could nevertheless be routed to the FM demodulator, which has a tuned circuit including a discrete coil, that has already been accounted for (Discrete Component Alert!). This coil could play a role in keeping the AM reception selective. It could be the signal oscillating in this coil, that gets AM-demodulated.
In order for this really to work however, the tuned circuit – i.e. the coil – of the FM demodulator, would need to be accurate to within 1%. This would mean that the production line is calibrating this coil – via machines – as each unit is being manufactured. I have my doubts that manufacturers still do that. Also, the coil would need to have a switchable Q-factor, which needs to be lower for FM reception, than it would then need to be, for AM.
(Update 7/9/2019, 16h55 : )
The assumptions I made with this second circuit are as follows:
- The supply voltage is shared with the first circuit, at 3V,
- The VCOs to be driven in any sort of PLL, have been computed to deliver their centre-frequency at Vcontrol = 2.4V.
- The desired amount of current was already flowing through R1 of the first circuit, which just happens to become 6μA,
- This notional amount of current is correct, even though M1 will only be conducting part of the time,
- VT0 and Vbias are both ‘close to’ 1.8V.
What seems to follow is, that the bias-voltage will be more negative from the supply voltage, twice as much as ‘Vcontrol-Nominal’, i.e., by 1.2V instead of by 0.6V. Therefore, R1 of the second diagram should be twice what it was in the first diagram.
There’s an added caveat. Even though the DC bias voltage for every mixer of this form, in a given device, may be the same, these transistors should be subdivided into ones that are to receive ‘high voltage-swings’ at their gate, such as ±0.5V, and ones that are only to receive faint, radio-frequency signals, which will always be M2. These groups of transistors could end up ‘talking to each other’, via their bias resistors. They should therefore each be connected to a separate instance of the second diagram from this posting.