4G/LTE chips coming . . . but first, a little chaos

SDR gets traction
"The SDR approach has gotten traction in LTE and WiMax designs," observed Joe Byrne, senior analyst at The Linley Group, Inc.

Besides CEVA, other DSP-IP suppliers such as Tensilica and NXP (CoolFlux) have landed customers for their DSPs, according to Byrne. "ST-Ericsson is using a DSP derived from an NXP design used in some of its 3G chips," he added.

There are technical reasons why industry players are pulling for SDR in the LTE basebands market.

"From a design standpoint, OFDM and MIMO processing are well suited to vector processing," said Byrne.

"OFDM, because its parallel transmission of data, and MIMO, because of its matrix operations." As data paths widen, the proportional overhead of a processor to control their sequencing goes down, nearing that of a purely hardwired solution, he explained.

Further, many believe that the immaturity of the LTE standard and ongoing development in MIMO processing increase the value of the flexible SDR approach.

"There's an argument that processors are easy to scale. To get from Category 3 LTE to Category 5, for example, a designer could add additional vector units (as Ceva-XC defines), additional processors in parallel, or operate the DSP at a faster rate," concluded Byrne.

But the argument to switch to a new DSP architecture should be much simpler. Handling in hardware different versions of radios deployed by various cellular network operators "gets too messy," CEVA CEO said. "You'd end up dealing with tons of hardware that need to be taped out. You can't possibly afford all the hardware spins it's just not sustainable."

Byrne agreed. "It's easier to validate a DSP-based chip, particularly if it uses a proven DSP, than to validate a full-custom chip. Designers can then whip the software into shape without having to iterate through the whole chip-design process."

Convincing vendors to go with a DSP-based SDR route for LTE/WiMax market is not a hard sell. In fact, as Strauss noted, all they'd have to do is to develop new DSP cores, likely multiprocessor ones, port their 3G stacks and develop LTE PHY and complete stacks for the new platform."

Moving off an in-house design

The challenge is "to move companies off an in-house design (be it hardwired or software-based) to a merchant solution," said Byrne.

ST-Ericsson's choice to use an in-house DSP surely dashed the hopes of merchant IP suppliers, Byrne observed. "We'll have to wait to see what other major baseband-chip suppliers choose to do. With the exception of Qualcomm, I expect most to license a DSP or acquire a small supplier using licensed IP."

Before moving from hardwired LTE to software, handset OEMs and semiconductor companies told CEVA that they first must see if CEVA's solution could meet certain conditions.

CEVA's CEO said, "One of the key criteria we kept hearing from different vendors was any solution that runs LTE Cat-4 (150Mbps DL, 50Mbps UL) needs to consume less than 250mW in order to remain competitive with first-generation hardwired solutions and meet their power consumption envelope." As for CEVA-X, Wertheizer noted, "We managed to get it to run at 150mW for both receive and transmit."