I see I have to explain. 'mosfet', also MOSFET => Metal Oxide Semiconductor Field Effect Transistor. More properly Insulated Gate Field Effect Transistor (IGFET) because metal oxide is just one of the things used for the gate insulator. Mosfets belong to the FET family.
There are two big families of transistor-type devices, junction-based and field effect. FETs are used for dense logic circuits because they can be made very small and the insulated gate variety draw power only when the logic state changes. Eack kind has its strengths and weaknesses, but the preponderance of logic circuitry means that many more FETS are manufactured.
BJTs (Bipolar Junction Transistors) operate by a current controlling a current, and their gain numbers are unitless. FETs are voltage controlling current, and their gain numbers have units of inverse resistance, which is conductance, so they are called transconductance devices.
I'm using a mix of them, trying to use each for its strengths. You can get the datasheets for them via Google: ksc1845, ksa992, lp0701, tn0680.
My device has multiple modules: voltage detector, timer, flasher (with three multivibrator stages), a low-voltage detector for the battery, and a pilot light module (the three-way pigeon flasher). There's also an input filter that faces the phone line, blocking the ringing signal, providing protection from voltage transients, and presenting a high impedance so that the box cannot affect the phone line. The output from the filter goes to the voltage level detector, which is concerned with detecting on-hook and off-hook voltages.
The voltage detector, the timer, and the control parts of the flasher operate on three volts, provided by three-terminal regulators (commodity semiconductor parts) from the battery voltage, which starts as high as 6.4 volts with new cells, and drops to about 3.8 volts. The undervolt detector (a commodity device with an adjustment provided by a BJT) turns on between 3.7 and 3.8 volts, and sends a signal to the 3-way pilot flasher to flash red, one very brief pulse roughly every six seconds. The box as a whole will keep working to a bit under three volts, but the batteries should be replaced long before then. It should be able to run the red flasher and the other circuitry for weeks before it stops working.
The voltage detector sends the off-hook signal to the timer, but it also sends off-hook and on-hook signals to the pilot flasher, causing it to flash yellow or green roughly every 20 seconds. The rate will depends a little on the battery voltage, and also on which LED is being fired.
The problem is that the pigeon flasher is driven directly off the battery (to avoid wasting energy in the regulator) and the logic on and logic off levels it needs are not those provided by the voltage level detector.
Both the off-hook detector and the on-hook detector use a BJT as the first stage, surrounded by resistors that bypass them until a certain current (measured in the nano-amps to micro-amps) makes it through the high-impedence input filter (DC impedance over 30 million ohms). What follows involves going from first-order approximations in theory to second-order approximations.
The low-level (off-hook) detector (which is bypassed and shut down when the high-level detector goes active) turns on and draws current from the positive rail through a resistor, causing the voltage between the resistor and the BJT to drop away from the positive rail. This turns on a MOSFET that (a) drives a hysteresis circuit and (b) operates another MOSFET to invert the logic and drive the timer.
Now, I may be able to drive that resistor from the unregulated (battery voltage) rail instead, if two conditions hold. First, the MOSFET has to be able to stand the gate going more positive than the source terminal. It can, within limits. I would be within those limits. Second, the voltage drop would have to be enough to turn the MOSFET on in spite of the extra voltage to be overcome, and that extra voltage would effect the circuit's turn-on voltage to some degree. Right now the resistor is 910 kilohms (all resistors having 5% tolerance in their values) and the turn-on current is about a third of a microamp, which means that I need about a three quarters of a nanoamp at the base of the first transistor (assuming a gain of 450 in the transistor at those current levels--see the datasheet). Its base-emitter bypass resistor is about 7 megohms, and at those current levels the base-emitter turn-on voltage is around four-tenths of a volt, so it needs about 55 nanoamps before it turns on.
Now, if instead of .4 volts to turn on the second-stage MOSFET, I need that .4 volts plust the possible 3.4 volts between the battery rail and the regulated rail, I will need about eight times the current at both output and input of the first BJT--that's six to seven nanoamps instead of three quarters after the 55 nanoamp bypass, a change from less than one percent to about ten percent, and -that- will change the input voltage threshold by the same ratio.
It appears that I can raise that 910Kohm resistor by a factor of five or so, reducing the current needed from the first stage transistor proportionally, and thus reducing the input current by nearly the same proportion. (Refer to the datasheet; at these collector (output) currents, the gain drops as the current drops.) This will reduce (but not eliminate) the load on the current through the input bypass resistor, and reduce the sensitivity to battery voltage.
The question to answer is what third-order effects occur. At lower battery voltages, the input transistor will be saturated, which will reduce the effect of the hysteresis circuit and might make the stage resist turning on at all. The quickest way to find out is to try it.
The added resistance will slow the turn-off of the detector output when the BJT turns off. The output-side MOSFET's insulated gate input acts like a capacitor, storing charge (but not so linearly; see the datasheet) and must discharge through the resistor. A larger resistor means a slower discharge, although the driving voltage will be higher. So long as I'm under two hundred milliseconds, I figure I'm fine. I can even tolerate a one-second with the box in use, but such delays make it very hard to measure the thresholds.
The high-level (on-hook) detector has similar issues, but they are magnified by the use of two BJT current amplifiers, and a third to bypass and shut off the low-voltage detector. I'm not going to drag that out here unless you insist, but there are also interactions with the low-voltage detector to consider.
