Data Converters in mmWave Systems

I was recently talking to an expert in phased array systems – someone who has worked on arrays for decades – and he said something interesting. To paraphrase, just like the cell phones of today bear little resemblance to the first brick phones, the mmWave systems 10 years from now will look very different from those prevalent today. I promptly took the bait and indulged in an exercise of day dreaming to imagine what may change in 10 years.

Most commercial millimeter wave (technically > 30 GHz) phased array systems of today employ analog beamforming. As illustrated in a receiver in the figure below, analog beamforming means that the phase shift is implemented at RF.

Here, the signal is phase shifted at RF and beam forming is obtained by controlling the phase shift of each output. There is another option to implement the phase shift – digital beam forming (see figure below).

This is not a new idea and has already been implemented in 802.11 systems (albeit with a smaller number of antennae) and radar systems. This does have numerous implementation challenges; however, since this is where I see commercial mmWave systems evolving to – let’s look at this in more detail.

As is obvious from the receiver in the figure above, the required phase shift is implemented in DSP after the data converters (i.e., in the digital domain). This has several inherent advantages and a lot of problems to be solved. Lets first look at the advantages (glass half full!). 

First, since the complex weights are implemented digitally, we can – at run-time – decide how many beams we would like to transmit. For instance, in a 64-element array, we can change the number of beams from 1 to 2 to 4 and so on. This allows tremendous flexibility when communicating with multiple targets (figure below). We can also digitally shape the beam and nulls where we want them (beam-steering and null-steering). For instance, in a receiver, if we know that a certain direction has a large blocker, then we can shape the null there (figure below). This is much easier to do in the digital domain rather than the analog one.

Second, note that the phased array is actually a subset of a timed array. The phased array works since, over a sufficiently narrow bandwidth, a phase shift is equivalent to a time shift. However, as bandwidths increase, the phased array is not sufficient since the error between a fixed phase and a time shift grows (resulting in the well-known beam-squint problem)

We can calculate the allowed bandwidth when using phased arrays – this results in, for instance, at 60 GHz, a maximum bandwidth of 240 MHz for a scan range of 60 degrees and a 256-element array. By implementing the ‘phase shifts’ in the digital domain, we can now allow for frequency dependent phase shifts which are effectively time shifts over a broader bandwidth. 

Third, since the phase shifters are implemented in DSP, the accuracy of the complex weights is digitally defined – i.e., it is as good as the digital resolution chosen for the specific implementation. No calibration is required and no drift over VT (voltage & temperature) is experienced.

With so many advantages, it would seem that digital beamforming is an obvious choice when implementing mmWave phased array systems. Unfortunately, we now come to “why this is hard”! 

Since each RF path is digitized, each path requires its own data converter. Thus, one ADC and DAC is required for each antenna – so, a 64 element array requires 64 sets of ADCs and DACs. Not only is this area hungry – but this is also very power hungry. Just to get an idea of the power consumption, a state of the art ADC with the appropriate dynamic range and sampling speed requires about 0.5-1W. This translates to ~32-64W for the array – just for the data converters! This is quite a thermal challenge – and on the surface, it looks like an architectural change in the data converter design is required in order to reduce these powers to a more manageable level.

In addition, another big challenge is that the entire receiver sees the wanted signal and all blockers. In a traditional analog phased array, this is not the case – the large off-angle blockers are cancelled once the signal summing is performed after phase shifting. Therefore, the dynamic range of the data converters necessary in digital beamforming is increased even more.

Even though mmWave systems are ostensibly about high frequencies, it looks like data converters are going to be the key block to enable digital beamforming. I’ll be keeping a close eye on how ADC and DAC designs evolve to meet this challenge!