The other day, I received an interesting e-mail from a reader regarding the selection of optical transceivers. I'd like to share it with you:
"I have a positive/pseudo emitter-coupled logic (PECL) signal running at 622 Mbps on an asynchronous-transfer-mode (ATM) interface board. I would like to have the option of selecting two kinds of optical transceivers. The transceivers are in different kinds of packages (Figure 1). Only one transceiver will be populated in the board at a time. When one position is populated, will the 622-Mbps signal see the open-circuited traces as some kind of loading, or will they appear as a true open circuit? Would a daisy-chain work better?" —Sek Ming Chua
The configuration in Figure 1 works only when the dimensions, or delay, of each leg are less than one-sixth of the signal rise/fall time (preferably one-tenth). In this case, the signal rise/fall time for 622-Mbyte/sec ATM is about 500 psec. One-sixth of that amount is 83 psec, which represents a trace length (at 180 psec/in.) of only about ½ in. It is doubtful that Sek Ming Chua will be able to reach between the two connectors without exceeding this distance. A daisy-chain configuration won't work any better. One way or the other, Sek Ming Chua will end up with an unterminated stub that is longer than ½in.
Let me quantify this discussion with a formula. If you ever had a class in transmission-line behavior, you may recall that the input-impedance magnitude of an open-circuited transmission line when excited by a sinusoidal waveform at frequency, f, is:
Z(f) = Z0/tan(2πfT),
where Z(f) is the impedance magnitude in ohms, Z0 is the characteristic impedance of the line, and T is the one-way delay of the transmission line.
You can use this formula to predict whether a short transmission-line stub will affect your signal. For example, in Figure 1, suppose you install Option 1 and leave Option 2 unpopulated. The main path from the transceiver logic to the Option 1 transceiver comprises trace segments A and B. Along this path, Segment C appears as an open-circuited stub.
For good signal quality, the impedance of Stub C should be very high compared with the Z0 of the main transmission path (A to B). In mathematical terms, you want the denominator of the above equation to be very small at all frequencies within the bandwidth of your digital signal. Assuming a rise/fall time of 500 psec with Gaussian slopes, the 6-dB bandwidth equals approximately f=½/(500 psec)=1 GHz. Further, assume that the length of stub C is about ½in. with a delay (T) of 90 psec and a characteristic impedance (Z0) of 50Ω. With these assumptions, the input impedance of Stub C works out to 79Ω.
This impedance significantly affects the transmission quality of Path A to B. My rule of thumb: Whenever the argument 2πfT exceeds one-tenth, I whip out my Spice simulator to see precisely what will happen in the time domain.
Sek Ming Chua needs some extremely small, zero-cost, high-performance switches. You can make such switches from ordinary solder pads and solder paste. Each switch is composed of one small pad with an approximately 0.030-in. diameter (round or square). The pad is split down the center with a gap of about 6 mils. If you put solder paste on the split pad at the time of assembly, the solder paste will short out the two halves and close the switch. If you don't apply any solder paste, the switch will stay open. Sek Ming Chua can use these assembly-time switches to connect the transceiver logic to one option or the other.