Here is a picture and circuit of a dusk switch I acquired recently. What is intersting is that this device is descrete...i.e. it uses no integrated circuits.
Until relatively recently most dusk switches used here in Tasmania were of the thermo-electric sort...basically a large, low impedance CdS cell as the sensing element (hence the large window in the side of the housing, which is the same shape of housing used for the older model with the CdS cell). This was plzaced in series with a 4K7, 5W resistor across the supply and the resistor was thermally coupled to a bi-metallic leaf which closed the contacts, turning on the load when the CdS cell wemt high impedance in the dark and the resistor cooled down.
Later, an electronic version appeared, it used a photodiode as the sensor, which fed its signal to a proprietary 8 pin DIL chip which housed the current-to voltage converter, low pass filters, latching flip flop and relay driver. A few external components "hung" from the pins of the chip, but all in all not much could be gleaned from it as to just how it functioned.
This circuit, however is discrete ant its function and design are fully exposed. Whoever designed this is a VERY CLEVER cookie! 2N7000 enhancement mode discrete MOSFETS are not often seen anywhere, but this circuit utilizes two of them. Like thermionic triodes, these allow for the design of a veyy high impedance front end. It is assumed that the front end...basically a Darlington type arrangement with the sensing photodiode or phototransistor, (as it was designated "Q1" in the board silkscreen) is "piggybacked" onto the first 2N7000. As the impedance of sich a design is so high the small 100nF bypass cap possibly acts as a full filter capacitor and the front end gets relatively smooth DC. This front end "Darlington" drives a second 2N7000 via a low pass network consisting id a diode, 470nF cap and 15 Meg resistor. This delays turn off and mitigates against nuisance tripping when dark clouds pass over during the day. I 1nF capacitor passes a small amount of ripple down to this circuit, the reason of which is not yet clear to me. This second 2N7000 has a common 2N3906 PNP in its drain circuit strapped by a relatively large 1.5uF monoblock capacitor. It is currently assumed that enough ripple passes via this capacitor to the drain of the 2N7000, which is configured as a source follower buffer driving the base of the real driver transistor, a 400v, 300mA KSP44 NPN transistor. Ripple, or more succinctly, lack of it at the collector of this final transistor is "rectified" by D5 turning on the 2N3906 (as it is supplied from the very large ripple rail "behind" D1, and this latches it "on"....well it latches the KSP44 on, but as the driven relay has NORMALLY CLOSED contacts, the load is turned OFF. So all day the relay is energized to keep the load off, at dusk the first 2N700, Q2, turns off slowly, the 470n cap, C4 slowly discharges via D7 and Q4, the second 2N7000 goes off too removing current from the base of the KSP44 and causing the 2N3906 to unlatch and the relay coil to be de-energized, closing the contacts and turning on the load.
In most electronics designs, ripple on a power rail is an "enemy" to be eliminated as mu\\ch as possible, but here, a very clever designer has actually designed the circuit around it and normal function requires it to get enough coupling via C5 to cause latching at dawn.
The use of 2N7000's harkens back to the days of thermionic valves, (Vacuum Tubes for our North Anerican members) where all sorts of clever stuff was done at very high impedances (and hig voltages).
What is interesting here though is how the maximum drain-source voltage of the 2N7000, which is 60v, is not exceeded, particularly at night where the circuit presents almost no load so the rail voltage will be high.
Another interesting point is the relay has a 125v A.C. coilfitted with a heavy copper pole shde. Admittedly it is running effectively of half wave rectified mains, rectified by D7.
The last point of note is its very wide operating voltage range of 105-305v. It seems that only the dark leakage current of the phototransistor is what saves the skins of the 2N7000's at the upper end of this range.
Sometime I will perform voltage and waveform measurements upon the operating device to glean more.