[time-nuts] PDIP package 100 MHz decade dividers

[ed breya]

One thing to be wary of, especially with 74AC running at full voltage, 
is that if you analog-bias an input midway, the internal supply current 
can go up. If the AC coupled input drive signal is big enough, then the 
input will spend little time near the middle threshold, so it's OK. If 
the drive is too small, or absent, then it will sit in analog mode, with 
higher power dissipation. The worst is say, in a self-biased inverter 
stage or string for squaring up a small RF input, where the output 
driver(s) also sits near the midpoint, if excitation is lost. In 74HC, 
this can get quite warm and severely load the power supply - especially 
problematic if you're depending on low power demand like in battery 
circuits. In 74AC, this can dump the power supply, or lead to self 
destruction from excessive power dissipation.

In systems where sufficient excitation is always assured, it's no 
problem to do the bias and AC coupling, but for applications where an 
input signal may be absent or too small, or has a duty cycle far away 
from 50 %, you have to watch out for the quiescent current situation. An 
example of this is a front panel RF input that goes into a 74AC 
square-up circuit.

There are simple tricks that can be used to ensure that the 74AC part 
can't sit for long in analog mode, like taking the presence or absence 
of a downstream signal for setting the bias at the front, which can be 
forced to one of the rails and cut off. These depend though on having a 
reasonably large input signal that can cross the input logic threshold, 
to get started. The input signal through the coupling cap will 
automatically rectify and push the logic input into range, by virtue of 
the internal protection diodes, or external clamps to the rails. 
External is better, since you have full control of the device types, and 
also, you can't be sure of the exact internal clamping structure. I 
always use small Schottky diode clamps for this, and always have some 
series damping/surge resistance - never run an external, uncontrolled 
signal straight into a 74AC input. The extra protection adds higher 
immunity to ESD and excessive signal power, pretty much eliminating any 
chance of crowbaring the device. 74HC and 74AC are already pretty tough 
in this respect, able to take 10 mA or so input overdrive (as I recall), 
without crowbaring. A little extra up front makes them tougher yet. 
Again, all this is for externally applied signals - inside of a system 
it's not necessary.

An alternative to controlling the front bias, is to control the power 
supply. This is my favorite method, and quite simple. The most recent 
example, is when I needed to make 30 MHz from 10 MHz, and used a 74AC04 
as an RF power amplifier/clipper. I set up a self-biased 3-inverter 
string for lots of gain, and used the remaining three paralleled and 
ballasted as the power output stage. A small 10 MHz reference is boosted 
to a good strong square wave with nice sharp edges. The result goes 
through a 30 MHz narrow BPF to separate it out.

This all would never work directly powered from 5V, but changing it to 
current-feed does. I fed it through a resistor from a higher supply (+15 
or +20 I think), and clamped Vdd at 5.1V with a Zener. This is not very 
efficient, of course, but it is simple. Other methods of current control 
would be just fine.

If there's no input, the AC7404 self-biases to some low supply (around 
1.5-2V I think) where the Vt of the devices starts to cut them off, by 
pulling enough current from the resistor. It remains in a linear range 
and active. When the signal is applied, it rectifies and amplifies and 
starts up the system, as described earlier, lifting everything toward 
normal operation. As most of the inverters start running full-swing, 
less quiescent current is needed by the IC, so the Vdd rises until the 
Zener takes over, holding it steady. Most of the current from the 
resistor is then used to develop the output power.

One thing I just recalled while writing, is that it was sometimes prone 
to self-oscillation without an input signal (just like many ECL 
prescalers), so that's a possible problem. I can't recall if I fixed 
that characteristic, or just ignored it, since it didn't matter in this 
application. If I did fix it, the solution probably would have involved 
reducing the HF open-loop gain, and cleaning up the circuit (combo board 
and air-wiring) layout.

Ed