The ConnX BBE16, from Tensilica, is a second generation baseband engine for LTE (long-term evolution) and 4G baseband SOC (system-on-chip) designs. The ConnX BBE16 is an ultra-high performance 16-MAC fixed-point DSP engine. It is based on an 8-way SIMD (Single Instruction, Multiple Data), 3-issue VLIW (Very Long Instruction Word) architecture with two 128-bit load/store units. It is a click-box configuration option with the configurable Xtensa LX3 processor core. Designers can also choose from a number of other configuration options (memories, interfaces, etc.) when designing their core. The ConnX BBE16 and an evaluation kit will be available in the second quarter of 2010.
The ConnX BBE16 can achieve 17 GB/s memory bandwidth at 550 MHz. Other performance metrics include: 16 multiply adds per cycle; four complex FIR taps per cycle; and one Radix-4 FFT butterfly per cycle. The improvements result in a 3x speed up on small matrix multiplications, a 20 percent speed up on complex FIR filter performance, and 2x speed up on polynomial approximations.
The ConnX BBE16 is fully programmable in C, and comes with an optimizing compiler and software tool set. Easy programmability, including automatic vectorization for ANSI C programs and optimized instructions for fast complex FFT, FIR, IIR, and complex matrix operations, make this new architecture ideal for quick implementation. To further speed the programming effort, optimized DSP kernel libraries are available for the ConnX BBE16 with functions ranging from generic DSP functions to specialized DSP kernels.
ConnX BBE16 DSP now supports:
- Matrix multiplication
- Real and complex FIR filters, including both symmetric and non-symmetric filters
- Polynomial approximations, especially for trigonometric and transcendental functions
- Key wireless functions such as polynomial generation and multiply-accumulate for de-spreading functions (up to 16 complex code MACs/cycle)
- High precision FFTs with adaptive range management
- Expanded support for vector normalization, shifts and packing
ConnX BBE16′s 16-way MAC (multiply accumulator) architecture is optimized for the demanding wireless DSP (digital signal processing) tasks, including OFDM (Orthogonal Frequency-Division Multiplexing) algorithms and FFT (Fast Fourier Transform), FIR (Finite Impulse Response), IIR (Infinite Impulse Response), and matrix computation. It is fully optimized for size and power-sensitive applications, and provides up to three times the performance of the original ConnX BBE on some of these critical algorithms. This architecture is particularly suited for system-on-chip (SOC) designs for programmable radio handsets, low-cost femto-cell and pico-cell base stations, micro and macro base stations, digital media broadcast receivers, and multi-format mobile DTV (digital television) demodulation.
More information: Tensilica