In this post, I will show an evolution of a trans-conducting LNA (rather than a voltage-gain LNA). This is a prime example of current-mode circuit design, which has benefits in terms of linearity—especially for low-voltage scaling in RFCMOS design.

## Conventional common-gate LNA

Consider the conventional common-gate LNA on its own:

This LNA consists of an input at the source of the MOS transistor (or emitter of a BJT) , a current-source to bias the transistor, and a resistor load. This configuration is typically used in broadband receivers (cable tuners, TV tuners, software-defined radio).

The gain of this stage is Av = gm×R1. This seems like a reasonable stage to produce a gain and therefore overcome noise.

Each of these circuits is presented as a single-ended version. However, there is absolutely no reason not to implement them as differential circuits. I merely draw them single-ended here to get the point across without obfuscating the primary ideas. If there is interest in seeing what the differential versions look like, please post a request in the comments section.

## CMOS LNM employing voltage-mode LNA

However, consider what happens when we couple this LNA into a CMOS switch-mode mixer:

The ⊗ symbol represents a set of commutating MOS mixers, typically capacitively coupled, to separate the dc bias voltages at the output of the LNA and the input of the post-mixer amplifier (PMA). You can essentially ignore then in this analysis, and just assume that they ideally convert from RF to baseband (direct conversion).

Consider the loss term R1/(R1 + R2). It doesn’t have to be there.

## Trans-conducting LNA

If we replace R1 with a current-source load and merely omit R2, we’d have something like this:

You’ll note that the previous loss term is now gone. We have greatly enhanced the gain at essentially no cost. Furthermore, the resulting architecture is even more linear: there is little swing at the output of the LNA; the summing junction of the PMA linearizes the LNA by presenting a low impedance to the LNA.

Due to the high output impedance of the LNA, a common-mode feedback circuit is necessary. I have detailed two ways of doing that.

## References

Matt Miller and I came up with these ideas at Motorola in 2004. We were not the only ones. Other people within Motorola came up with the same idea. In addition, I had seen it published somewhere around 2004—I thought by Michiel Steyaert or Thomas Lee. Despite my recollection, the closest references I have seen available are: “A 72mW CMOS 802.11a direct conversion receiver with 3.5dB NF and 200kHz 1/f noise corner” (albeit using an inductor load instead of PMOS loads) and “A 1 V 1.1 GHz CMOS integrated receiver front-end” (with a folded Gilbert cell mixer).