To make the transition from 3G to 4G technologies, wireless baseband vendors face a huge increase in processing requirements, due to both the greater complexity of the underlying orthogonal frequency-division multiplexing (OFDM) algorithms and the increase in data rates. At the same time, ongoing changes in 4G protocols such as WiMAX and Long-Term Evolution (LTE) create a need for flexibility.
To meet these needs, most initial 4G basebands use a soft-modem approach that combines multiple programmable CPUs and DSPs. This approach is a change from 3G modems, which tend to rely mainly on hardware accelerators. The soft approach has some problems, however, that may eventually push the pendulum back in the other direction.
2G To 3G To 4G…
Whereas 2G basebands can be implemented in software on a small DSP, 3G protocols require an increase in both DSP performance and data rate. As a result, most 3G baseband processors use hardwired accelerators to offload the most intensive signal processing from the DSP, allowing the programmable unit to focus on 2G compatibility. A separate CPU is often added for protocol processing.
The development of 4G basebands has caused a surge of interest in an approach called software-defined radio (SDR). This method uses a collection of programmable DSPs and CPUs to implement the entire baseband function. In theory, an SDR baseband can be programmed to perform any wireless protocol.
This capability could enable a single baseband to support roaming from LTE to WiMAX networks, for example, and backward compatibility with existing 2G and 3G networks. In addition, the level of programmability provides the ultimate in flexibility, accommodating ongoing standards development.
In practice, SDR has two problems. The first is the extensive programming required to support so many protocols, particularly considering that other solutions already exist for 2G and 3G. Some vendors have decided that it is simpler to add their current, proven baseband circuitry for 2G or even 3G. At 40 nm and below, the die area of a separate 2G baseband is tiny, and a 3G baseband is not much bigger.
The second problem is power. As a general rule, a hardwired design uses half the power (or less) of a soft solution. Even SDR proponents acknowledge that their approach requires higher power. In some cases, the performance of the SDR baseband must be throttled back to fit within mobile power requirements, limiting its performance.
Therefore, most commercial 4G baseband designs add some level of hardware acceleration to the SDR core. Commonly used functions such as fast Fourier transform (FFT) and forward error correction (FEC) are obvious candidates for hardware acceleration. Other portions of the OFDM algorithm can also be cast in hardware, because this algorithm is common to all 4G protocols.
Some vendors like to use the SDR term and others do not, but we see the various implementations forming a continuum rather than segregating into two camps. None uses a fully programmable design, but none is fully hardwired, either.
Evolving The 4G Solution
The initial LTE and WiMAX devices are almost exclusively USB dongles for laptop computers. Only a few 4G smart-phone models are available. Because USB dongles are powered from the laptop’s large battery and stick out from the body of the computer, they have relatively little restriction in power consumption, heat dissipation, and size.
Continue on next page
Therefore, most device vendors aren’t concerned that the initial 4G baseband processors are bulky and burn far more power than common 3G designs. These processors typically require a separate chip for 2G/3G compatibility plus their own RF subsystem and sometimes additional glue logic as well.
Naturally, they cost much more than 3G processors as well. Like a dancing bear, these chips (or chipsets) are impressive not because they do 4G well, but because they do it at all.
As the demand for 4G increases, however, chip suppliers will be under more pressure to reduce the size, power, and cost of their designs. Moving to 28 nm and beyond will provide some help, particularly on the size and cost fronts. But as chip designers know, advanced IC processes are no panacea for power problems.
For these reasons, we expect 4G baseband designers to move more functions from software to hardware over time. Now that Release 9 has solidified the LTE standard, hardwiring some of the protocol becomes feasible. At data rates of tens of megabits per second, processing the data in software burns too much power.
Hardening these functions will require extra transistors, particularly in processors that support multiple protocols. But as noted above, transistors continue to get cheaper over time. Trading more transistors for less power is the right long-term approach. Design and validation is more expensive and time-consuming for hardwired circuitry. But as 4G volumes grow, amortizing these costs becomes much easier.
The first generation of soft 4G modems represents the infancy of LTE basebands. As cost and power pressures increase, they will evolve to use more hardware acceleration, like today’s mature 3G baseband designs. As 4G roadmaps push toward even greater performance, power will be the critical limiter, driving basebands to use less software and more hardware.