Ambarella (United States)

Genetic algorithms are robust parallel calculation methods based on natural selection. Various crossover and mutation methods to accomplish Genetic Algorithm (GA), namely, single point, multipoint, uniform, greedy, migration, and on-demand etc.; exist. However, these mechanisms are static in nature. This paper presents a dynamic crossover (DC) mechanism. We investigate its performance by implementing in hardware (FPGA) with convergence rate and higher fitness as the performance metric. The purpose of the DC concept is two fold; to achieve faster convergence and to consume lesser memory by keeping the population size static. The results indicate that for a linear and a nonlinear objective function, DC outperforms all static crossover mechanisms.

Video transmitted over fading wireless channels incurs both encoder-induced distortion and distortion caused by transmission errors. To address these problems, we propose a framework to transmit video reliably over a multi-antenna wireless system. No assumptions are made about specific video-coding standards, except that the encoder employs some variation of motion-compensated block-transform coding. To minimize the total distortion, the scheme optimizes the recently- derived diversity and multiplexing trade-off available on multi-antenna systems. Moreover, this is done jointly with loss-aware rate-distortion optimization (LA-RDO) techniques at the video encoder. Besides minimizing distortion, the algorithm also ensures that delay and buffer constraints are satisfied for real-time transport. We show that problem can be modelled as a short sequence of geometric programs. The overall complexity is linear in the number of antennas. Numerical results with H.264-coded video show that the PSNR at the video-decoder can be improved by various amounts, depending on antenna configuration.