Example of a Transimpedance Amplifier

Example of a Transimpedance Amplifier

Transimpedance amplifiers are often used as photodiode preamplifiers in optical communication links. They are also used as a component in certain wideband, fast transient-response amplifiers such as the Cherry–Hooper design [14]. One of the simplest ways to implement a transimpedance amplifier is with the dual feedback (series–series and shunt–shunt) approach shown in Figure 74.12. This design has the capability of using feedback to not only exchange gain for bandwidth but also to provide a wideband low-impedance match at the input and output. For this example, let us suppose that the amplifier is designed to have equal input and output impedance and a transimpedance gain of 6 dB. An InP HBT device is used to produce realistic simulations and time constant numbers for this example. ZIN and ZOUT were 50 W, and a bias current of 10 mA was selected.

The design equations for the voltage gain, Av and input and output impedances of this amplifier are shown below.

The same procedure as described in Sections 74.1 and 74.2 was used to find the time constants. In this case, the design is small signal, so the device transconductance and capacitances did not require linearization. Once the bias voltages and currents are determined, these elements can be calculated and their respective time constants found. The device model was simplified to include only CBE, CD ,

and CCB capacitances. Rbb was lumped as one resistor and

RE = RE + REX. The contribution of the depletion and diffusion base–emitter capacitance were combined as Cp = CBE + CD. To simplify the calculations further, the transconductance, gm, and Cp were degenerated as shown in Eq. (74.41) and Eq. (74.42) so that the emitter feedback resistor is absorbed, and a common-emitter equivalent circuit as shown in Figure 74.13 is produced. This eliminates the difficulties in analysis that the series feedback otherwise produces.

Table 74.8 shows the two main OC time constants and the SC time constant derived for this design and evaluated using small-signal model parameters for the InP HBT. The b1 and b2 coefficients that were calculated from these time constants are also shown in this table. The pole frequencies estimated by Eq. (74.38), Eq. (74.45), and Eq. (74.46) are found to be 29 and 401 GHz. Thus, the widely separated pole approximation is valid, and since a dominant pole is present, one would expect a good, well-damped transient response for this amplifier with little or no gain peaking in the frequency response. Table 74.9 shows the wn, z, and 3 dB bandwidth predicted by the equations in Table 74.8. Since z > 1, we do not have complex-conjugate poles, so the method should provide reasonably good estimates. The simulated frequency response of the same amplifier is shown in Figure 74.15. The simulated bandwidth is about 30 GHz, in reasonable agreement with the estimate.

In addition, in data transmission applications such as optical communication, the transient response is also very critical, more so than the bandwidth. The best pulse response requires a constant group delay over the required signal bandwidth. A model of the source is also important to include in transient simulations, because photodiode capacitance and bondwire inductances will affect the response. Figure 74.16 illustrates one such model obtained from the literature [15]. A linear feedback shift register source is used to simulate random data. Figure 74.17 shows the output voltage of the amplifier presented as an eye diagram for this source and amplifier combination. There is a little overshoot caused mainly by the bondwires and no ISI is evident.

The performance seems adequate for a 40 Gb/s NRZ application. But can the performance be improved further to possibly use this technology for a RZ application which would require higher

bandwidth? Both the OCTCs are about the same, but if one of them could be reduced, a possible improvement in bandwidth and risetime could be obtained. A Darlington implementation as in Figure 74.18 comes to mind since the low output impedance of the emitter follower could reduce the contributor. One risk owing to the emitter follower could be ringing due to the inductive output impedance. Additional damping may be required. Rather than deriving all of the time constant predic- tions for this second case, which would be similar to the clock input part of the ECL SFD analysis in Section 74.2, the simulation of the frequency response is shown in Figure 74.19. A 50% increase in 3 dB bandwidth was achieved by the addition of the emitter follower at the cost of gain peaking. The higher bandwidth would seem to be very promising for higher data rate applications; however, the transient response must also be examined. Gain peaking is a sign of an underdamped transient response. Figure 74.20 shows the result of the same transient analysis that was used for the unmodified 30 GHz bandwidth amplifier. In this figure, the ringing is quite evident, and ISI has definitely increased to the point where the amplifier is probably useless for this application, even at 40 Gb/s. Additional modifi- cations could be incorporated to possibly damp the emitter follower ringing. Further design work is needed to make this amplifier useful. Thus, bandwidth alone is not a sufficient measure of performance for pulse applications.