Architecture optimisation Optimal Determination of Common for SDR C. Moy, J. Operators for Multi-Standard Software Palicot, Virgilio RODRIGUEZ, D. Giri Defined Radio Overview 4th Karlsruhe Workshop on Software Radios Mathematical Karlsruhe, Germany framework Architecture optimisation Discussion / C. Moy J. Palicot Virgilio RODRIGUEZ D. Giri Outlook SCEE Group, IETR/Supélec Cesson-Sevigné, France email: vr@ieee.org 23 March, 2006
Outline Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio Overview 1 RODRIGUEZ, D. Giri Overview 2 Mathematical framework Mathematical framework Architecture optimisation 3 Architecture optimisation Discussion / Outlook Discussion / Outlook 4
Problem statement Architecture The optimal design of a multistandard reconfigurable optimisation for SDR radio is the right choice between two extremes: C. Moy, J. One extreme: the “Velcro” solution (one self-contained Palicot, Virgilio RODRIGUEZ, complex module for each supported standard) D. Giri Other extreme: install only the most “primitive” Overview components (adders, multipliers, etc), and provide Mathematical ”higher level” functionality through multiple calls framework Trade-off: Architecture optimisation Velcro architecture generally provides best Discussion / performance, but at highest manufacturing cost (and Outlook size/weight) Other extreme likely minimises cost (& size/weight) but at unacceptable performance (multiple calls add latency!) Our approach : build a mathematical framework to find the optimum between these extremes
Overview Architecture We model the reconfigurable radio as a (hyper)graph of optimisation for SDR progressively simpler functional modules C. Moy, J. Palicot, Virgilio The functionality of a given module can be provided in RODRIGUEZ, D. Giri 2 ways: Overview installing a dedicated component optimised for that task Mathematical invoking (repetitively) lower level modules framework With each module we associate 2 “costs”: monetary Architecture optimisation and computational (delay) Discussion / Outlook When a lower-level module is needed several times it is invoked multiple times (not physically replicated ) The cost of a design is a weighted sum of the totals of both costs To find the optimum, we use: (1) exhaustive search & (2) simulated annealing
A graph for a tri-standard radio Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio RODRIGUEZ, D. Giri Overview Mathematical framework Architecture optimisation Discussion / Outlook
A realistic “sub-design” example Architecture Want an architecture to support 3 main functional optimisation for SDR modules: OFDM, Equalisation, and Multichannel C. Moy, J. Palicot, Virgilio processing RODRIGUEZ, D. Giri Presumably these modules are part of grander design Overview Equalisation (to compensate for multipath) can be Mathematical implemented via framework Architecture FIR filtering optimisation FFT (great for channels with long impulse responses) Discussion / Outlook Multichannel refers to channelisation function of BS (needs to process many channels in parallel). Two options: “Classical” channel per channel Filter bank channeliser (which can be implemented via FFT)
Sub-graph of design choices I Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio RODRIGUEZ, D. Giri Overview Mathematical framework Architecture optimisation Discussion / Outlook
Performing the optimisation Architecture optimisation Key question: should we install a for SDR self-contained/dedicated component to perform a given C. Moy, J. Palicot, Virgilio functionality, or should we invoke lower level RODRIGUEZ, D. Giri modules/components? Overview Each component is characterised by 2 “costs”: Mathematical monetary, and “computational” (time) framework Architecture When a lower-level module is needed several times it is optimisation invoked multiple times (not physically replicated ) Discussion / Outlook Choose components to minimise a weighted sum of total monetary plus total computational costs Algorithms: Exhaustive search (“brute force”) Simulated annealing
Sub-graph with some parameters Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio RODRIGUEZ, D. Giri Overview Mathematical framework Architecture optimisation Discussion / Outlook
Results Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio RODRIGUEZ, Results are heavily influenced by chosen weights D. Giri (monetary vs. computational) Overview when “delay” costs weigh heavily, complex, expensive Mathematical framework but high-performing dedicated components are chosen Architecture when “delay” costs weigh less, simpler, reusable optimisation components are chosen (leading to a less expensive Discussion / design but with higher latency) Outlook Above agrees with intuition
An optimal design Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio RODRIGUEZ, D. Giri Overview Mathematical framework Architecture optimisation Discussion / Outlook
Discussion Architecture optimisation We presented a mathematical framework to find an for SDR optimal architecture for a multistandard reconfigurable C. Moy, J. Palicot, Virgilio radio RODRIGUEZ, D. Giri Key: graph of progressively simpler functional modules, Overview showing their interdependencies (AND, OR ) Mathematical framework Key question: install (specialised component) or invoke Architecture (lower levels)? optimisation Choose components to minimise weighted sum of 2 Discussion / Outlook “costs”: money and delay A realistic “sub-design” has been solved both by “brute force” and by simulated annealing Results are highly influenced by weights, and are intuitive
Immediate Future (in progress) Architecture optimisation for SDR Re-building the hypergraph of design choices. C. Moy, J. Researchers seek: Palicot, Virgilio RODRIGUEZ, new operators that may be common to several D. Giri communication blocks Overview to replace time-domain with new frequency-domain Mathematical algorithms (which would add arcs pointing to FFT ) framework to include more communication standards in the graph, Architecture optimisation and track their evolution Discussion / Outlook Change objective function to minimise (monetary) cost only, subject to delay constraints (“deadlines”) Transform the architecture optimisation into a “network design problem” (to access extensive literature with many algorithms and results)
In the more distant horizon Architecture optimisation for SDR Consideration of: C. Moy, J. Palicot, Virgilio RODRIGUEZ, multiple instances of same component (butterfly, FFT, D. Giri etc) to reflect real market choices Overview time needed to re-configure the radio while switching Mathematical framework standards Architecture optimisation “travel time” of signals from a component to another Discussion / possible contention among high level modules for the Outlook service of the same lower-level module (which may be critical if the SDR needs to support simultaneous operation over several standards)
A glance into the future: graph/network Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio RODRIGUEZ, D. Giri Overview Mathematical framework Architecture optimisation Discussion / Outlook
A glance into the future: graph/network Architecture optimisation for SDR C. Moy, J. Palicot, Virgilio RODRIGUEZ, D. Giri 0/s OFDM 0/s ButterF FFT Overview 0/s Equalis 0/s €/0 Mathematical framework 0/s €/0 FIR Architecture €/0 €/0 optimisation origin CIC 0/s Discussion / €/0 Outlook €/0 Cordic €/0 Map 0/s ChPCh 0/s 0/s FiltBk MultiChan
Recommend
More recommend
