Research On The Technologies Of Low Cost Digital Signal Processing For Space Based Network Intelligent Satellite

A satellite,as an service node in the space based network system(SBNS),whose computing capacity and intelligence directly impacts the quality and the efficiency of the SBNS.Intelligent satellite is becoming the key of development of the SBNS due to its open architecture,universal hardware platform,open sourced software system and on orbit customized ability.Besause of the volume,weight and power limitation for the intelligent satellite,the capacity and resource of the computation are two standing out conflicting factors,the ability to reduce the cost of computation is becoming the key of signal processing for intelligent satellite.This thesis realised the balance between the capacity and cost for the real time signal processing.The contributions and novelty of the thesis are the following:1.To solve the need of addition for low cost real time signal procssing for the intelligent satellite,approximate adders are proposed for the massive computation,including a transmission gate based approximate gate level full adder and a logic reduction based optimal design methodology for component level adder.First of all,based on the principle of achieving complex logic function by using simple transmission gate,meanwhile in order to reduce the power and carry propagation chain,in this paper,two novel types of approximate adders are proposed based on the transmission gate.Compared to transistor reduction based approximate mirror adder,the two proposed approximate full adder cell could reduce 25.58% power,31.8% delay,49.12% power delay product and 14.45% power,31.5% delay,41.32 power delay product on average respectively,meanwhile reduce 16.44%,39.61% of the absolute error on average.which builds the foundation of designing component level approximate adders.In the image sharpening application,compared to other designs,the proposed design could provide high accuracy,i.e.,high quality image.Second,a logic reduction based optimal design methodology is proposed.The method initially forms the analytical model of mean squared error and mean absolute error for approximate adder.Based on the analytical model and aided by optimized logic synthesis,three low cost and high accuracy approximate adders are proposed.Compared to timing starving approximate adders,the proposed design could reduce 20.64%,29.28% and 19.83% power on average respectively,meanwhile reduce 49.25%,49.45% and 32.32% mean squared error on average,in others words,forming the Pareto optimal frontier.In the application of image compression,compared to other designs,the proposed design could substantially reduce the power while providing high image quality and high accuracy.2.To solve the need of multiplication for low cost real time signal procssing for the intelligent satellite,a compensation based design methodology for approximate multiplier is proposed for the massive computation.The method initially analyzes the fundamental principle of approximate multiplier which is essentially the compensation of truncation.Based on this,under the constraint of minimal mean squared error,a conditional based optimal multiplier design methodology is proposed for practical use.Compared to approximate compressor based multiplier,the proposed design could significantly reduce the power meanwhile providing a higher accuracy,in others words,forming the Pareto optimal frontier.In the application of FIR filter,compared to other designs,the proposed design could substantially reduce the power while providing high signal accuracy.3.To design approximate linear time invariant system for low cost signal processing,an optimal method based on the approximate constant multiplier is proposed.The approximate constant multiplier is based on the truncation of lower weight partial products.The relation of truncation length and the mean squared error and mean absolute error is mathematically modeled respectively.Based on the models,when given the systematical accuracy constraint,an optimal truncation length allocation method is proposed.In the application of IIR filter,compared to other designs,the proposed design could substantially reduce the power while providing high signal accuracy.All the research results above confirmed to be efficitive and practical with therotical analyses and expertimental verifications.They are of great theoritical and practical significance to the design of low power digital signal processing system for the space based network intelligent satellite.