A Software Radio Approach to Commercial FM Content Indexing Bruce Denby, Olivier Romain, Sid-Ahmed Hariti Laboratoire des Instruments et Systèmes d'Ile de France (LISIF) Université Pierre et Marie Curie B.C. 252, 4 place Jussieu, 75252 Paris Cedex 05, France email: denby@ieee.org Abstract. – The article presents a feasibility study for a Software Radio based commercial FM content indexing system, focussing on the possibility of future portable and embedded implementations. Keywords: broadcast monitoring; audio indexing; software radio; FM radio I. INTRODUCTION There is considerable interest today in the use of audio indexing techniques to facilitate classification and retrieval of information contained in commercial broadcast media [1-7]. Such technology is in demand both by the broadcast monitoring community - for intellectual property rights issues, and by the entertainment industry - for providing consumers with novel media search and retrieval services. Audio indexing may be content-based - for example speech/music discrimination based on the audio waveform - or metadata-based, exploiting explicit tags broadcast along with the audio content - such as in the Radio Data System (RDS) stream contained in most commercial FM radio broadcasts. Commercial broadcast monitoring systems are typically rather massive installations, due to the large number of channels to be treated. For many consumer applications, however, it would clearly be advantageous, to imagine portable audio indexing solutions - for example in conjunction with an FM car radio, to see if a particular song has been played in the past few minutes, or to tag station content in real time as voice or music. Recently [8], software radio has been proposed as a possible enabling technology for this purpose. Originally conceived to provide adaptable air-interfaces for the wireless telecommunications industry, software radio is now discussed for commercial FM radio applications as well [9,10]. In an audio indexing context, software radio is interesting in that it digitizes entire chunks of radio spectrum and analyzes them in software. Thus, coupled with a memory module or a hard disk, a software radio solution with sufficient processing power could provide a means of transforming the FM radio band into an indexed, searchable 'Hertzian Internet.' The programmability of software radio architectures, furthermore, would make them all the more attractive as future DAB and iBOC services - also foreseen to make extensive use of metatdata tags - begin to roll out. Suitably equipped receivers could take advantage of these new services without hardware upgrades, simply by downloading a new software module. The article presents the results of a feasibility study for a software radio based FM band 'navigator' intended to enable both content- and metadata-based audio indexing processes. The navigator concept is first introduced in section II. Section III then describes the hardware and algorithmic studies performed to assess the feasibility of a portable or embedded FM navigator solution. In the final section, a conclusion and a discussion of planned future developments are presented. II. THE SOFTWARE RADIO ENABLED FM BAND NAVIGATOR CONCEPT. The FM navigator concept is depicted in figure 1. After reception by the antenna and low noise amplifier (LNA), the downconverter and bandpass filter (BPF) move a segment of radio frequency spectrum - in this case, the entire 20 MHz-wide commercial FM radio band - down to baseband for digitizing. The downconvert is 'optional' in the sense that if a digitizer of sufficient speed is chosen, downconversion is unnecessary. This mode of operation is to be contrasted with that used in a standard FM radio, where a phase-locked loop (PLL) is used to downconvert and demodulate one station at a time. If all channels were desired in parallel, separate downconverters, filters, and digitizers would thus be necessary. In the software radio approach, it is not until