Accurately predict measured noise figures for transformer coupled differential amplifiers (Part 2 of 2)

Fixed transformer turns ratio designs using a range of FDA’s

Selecting a range of the highest performing voltage feedback FDA’s and fixing the turns ratio at 1:2, swept gain NF comparisons can be generated using equation 3. Recall that β is the estimated transformer insertion loss while α is the ratio of R_f to R_g in the intended application circuit of Figure 1 (repeated here). This simplified drawing is showing just a step up turns ratio (n, in equation 3) but not the attendant insertion loss that is also specified for wideband transformers.

Figure 1(repeated from Part 1 of this article):The intended application circuit is shown for this analysis.

Many modern FDA’s offer extremely wideband, low power solutions. For instance the ISL55210 is only 115mW giving a 4GHz gain bandwidth product. Table 5 shows 4 exceptional FDA’s with their data sheet noise and Gain Bandwidth Product (GBP) numbers. These are estimates in some cases and the 10GHz number for the LT6904 is correct for higher gains but external C compensation is required at lower gains to hold stability – reducing the useable bandwidth. That issue will be neglected here for simplicity.

Table 5. Table of FDA’s and key parameters is shown.

The turns ratio of 2 seemed to hit a good NF profile and will hold higher amplifier loop gain and bandwidth for any given total gain target. Putting each device’s spot noise numbers into equation 3 with the ADT4-6T insertion loss gives the relative noise figures of figure 5 swept up in target total gain.

Figure 5. Estimated NF vs total gain shown on graph.

The ISL55210 and ADA4930 provide nearly identical predicted results with the higher noise THS4509 and LTC6409 running somewhat higher.

Estimating the available bandwidth over gain

The overall solution bandwidth will be combination of the transformer bandwidth and the estimated bandwidth of the FDA. Sweeping the same 4 devices and estimating the amplifier bandwidth from their stated gain bandwidth product (GBP) and the noise gain (1+α/2) gives the curves of figure 6. While the LT6409 suggests very high bandwidth, for amplifier gains <= 20 it needs external C to hold stability which bandlimits it much lower than the straight gain bandwidth product calculation used here.

Figure 6. This graph shows estimated F-3dB in the amplifier swept across gain in the circuit of Figure 1 with a 1:2 step up.

Table 6 lists some representative wideband 1:2 turns ratio transformers with their insertion loss and low end to high end frequency cutoffs.

Table 6. Representative wideband pulse transformers are charted here with turns ratio = 2 (ohms ratio = 4)

While there is a considerable range of passband frequencies here, a good high end estimate for those with low insertion loss might be 800Mhz. Taking the amplifier bandwidths of figure 6 and RMS’ing with an 800MHz transformer bandwidth gives the net achievable signal bandwidth shown in figure 7.

Figure 7. The graph shows the result of combining the estimated amplifier bandwidths with a transformer

The ISL55210 offers the best combination of useable bandwidth over gain with the lowest Noise Figure and power dissipation in this circuit. Both the ADA and THS parts are slower, while the ADA device does match the NF closely for the ISL55210. The LTC device will give a circuit that is mainly transformer bandwidth limited but with a slightly higher NF profile.