doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 1 Project: IEEE P Working Group for Wireless Personal Area Networks (WPANs) Submission Title: [Complexity and Performance Analysis of a DS-CDMA UWB System] Date Submitted: [ 17 September, 2003] Source: [Jaiganesh Balakrishnan, Anand Dabak, Srinivas Lingam and Anuj Batra] Company [Texas Instruments] Address [12500 TI Blvd, Dallas, TX 75243, Texas, USA] Voice:[ ], FAX: [ ], Re: [] Abstract:[The following contribution provides a complexity and performance analysis of a DS-CDMA UWB System.] Purpose:[This document reflects the contents of xxxxx in a more presentation-friendly format.] Notice:This document has been prepared to assist the IEEE P It is offered as a basis for discussion and is not binding on the contributing individual(s) or organization(s). The material in this document is subject to change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or withdraw material contained herein. Release:The contributor acknowledges and accepts that this contribution becomes the property of IEEE and may be made publicly available by P

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 2 Complexity and Performance Analysis of a DS-CDMA UWB System Jaiganesh Balakrishnan, Anand Dabak, Srinivas Lingam and Anuj Batra Texas Instruments TI Blvd, MS 8649 Dallas, TX 75243

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 3 Multi-path for DS-CDMA Multi-path reflections from different objects in a room –Remember light travels at ~ 1 ns a foot –DS-CDMA chip period is 731 ps –MBOK symbol period is 17.5 ns –CM3 channel reflections easily up to 40 ns –There is cross-talk between the different multi-path reflections. Object 1 Object 2 Reflection 1 Reflection 2 Object 3 Reflection 3 MBOK code Cross talk between multiple reflections

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 4 Multi-path cross talk effect in DS-CDMA Simulations for DS-CDMA should take into account multi-path cross talk. Simplistic simulations for DS-CDMA can be done by just summing the energies from the different multi-paths –However, these simulations do not take into account cross-talk between the multi-paths –Will give incorrect and over optimistic results Reflection 1 Reflection 2 Reflection 3

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 5 Multi-path Performance In multi-path environments, the RMS delay spreads for a UWB channel can be large (14 ns for CM3, 25 ns for CM4). The performance of a DS-CDMA system in multi-path channel environments is determined by the following factors: Multi-path Energy Capture: –Function of the number of RAKE fingers used. –Uncollected multi-path results in lost energy ICI/ISI: –Degradation due to ICI/ISI depends on the severity of the channel, data rate(spreading gain) and the equalizer structure.

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 6 Multi-path Energy Capture (1) Assumptions: –Optimal timing information. –Perfect channel estimation. –Largest RAKE fingers over the entire span of the channel impulse response are selected. –No shadowing –Does not reflect degradation due to ICI/ISI. –Loss in captured energy averaged over all 100 channel realizations. Observations: –Average loss of 2.5 dB with a 16 finger RAKE for CM4 channel environment. –Average loss of 1.3 dB with a 16 finger RAKE for a CM3 channel environment.

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 7 Multi-path Energy Capture (2) The performance for the 90 th %-ile channel realization is the metric of interest [03/031, 03/276]. –03/031: Selection criteria document –03/276: CE SIG requirements Assumptions: –No Shadowing –Does not reflect degradation due to ICI/ISI. Observations: –90 th %-ile channel realization has a loss of 3.8 dB with a 16 finger RAKE for CM4 channel environment. –90 th %-ile channel realization has a loss of ~2 dB with a 16 finger RAKE for a CM3 channel environment.

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 8 Impact of ICI/ISI (1) Even if an infinite finger RAKE is used to collect all the multi-path energy, the DS-CDMA system will have a performance gap from AWGN due to the effect of ICI/ISI. Assumptions for the DS-CDMA System: –Chip rate of 1368 MHz. –SRRC pulse shape with 50% excess bandwidth. –R = 1/2, K = 7 convolutional code. –Information data rate of 114 Mbps a processing gain of 12. –No back-off at the transmitter due to ripples in the PSD. –CM3 channel environment. –No shadowing. –Perfect channel estimation. –Optimal finger placement for the RAKE, i.e., the locations with the largest channel taps are picked.

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 9 Impact of ICI/ISI (2) Assumptions for the OFDM system: –MB-OFDM system with 3-bands and a total information tone bandwidth of MHz. –R = 1/2, K = 7 convolutional code. –Maximum uncoded data rate of 640 Mbps with QPSK modulation. –Information data rate of ~110 Mbps a frequency domain spreading factor of 3. –No overhead due to the OFDM prefix. –CM3 channel environment. –No shadowing. –Perfect channel estimation. Observations: –Infinite finger RAKE has an average degradation of 0.7 dB over AWGN due to ICI/ISI. –OFDM system loses 1 dB over AWGN due to loss in frequency diversity. –A 16 finger RAKE has a 2 dB loss over AWGN.

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 10 Impact of ICI/ISI (3) What happens when we look at the 90 th %ile results rather than the average performance? Observations: –The infinite finger RAKE has a ~1.7 dB degradation over AWGN performance. –The OFDM system has the same 90 th %ile performance as the infinite finger RAKE. –The 16 finger RAKE has a ~4.5 dB degradation over AWGN and ~2.8 dB degradation over the OFDM system. MB-OFDM achieves infinite rake performance with implementable complexity

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 11 Range of DS-CDMA System Document [03/153] does not provide results for the 90% link success probability distance. Average range results are optimistic and does not provide the complete picture. Impact due to ICI/ISI for the 90 th %ile realization is nearly 1.7 dB for the 114 Mbps data rate. Need a margin of 3.8 dB to compensate for the impact of log-normal shadowing to obtain the range for the 90 th %ile link. * - Based on the results provided in document [03/153] RAKEAWGN* Loss in captured energy ICI/ISI + Shadowing Total degradation Expected Range Infinite21.6 m0 dB5.5 dB 11.4 m 16 fingers21.6 m2 dB5.5 dB7.5 dB9.1 m 5 fingers21.6 m4.5 dB5.5 dB10 dB6.8 m MB-OFDM has a 90% link success probability of ~11.5 m

doc.: IEEE /388r0 Submission September, 2003 Jaiganesh Balakrishnan et al., Texas InstrumentsSlide 12 Complexity comparison RAKE receiver for DSSS * –Computational cost of correlators Each Correlator requires 24 additions/subtractions at the chip-rate (e.g., 1.368GHz) For RAKE of length 16, need 8 x 16 = 128 correlators for 16-BOK => 175,000 Madds per second !! –Computational cost of MRC (this occurs at symbol rate. E.g., 57MHz) 16 vectors of length 8 need to be multiplied by the corresponding 16 complex conjugate channel tap weights (one complex tap weight per vector). 16 x 8 x 57 = 7296 MOPS (non-trivial complex multiplies) Overall 7296 M complex multiplies M complex adds/sec!!) Does even not include multi-path equalization complexity FFT for OFDM (MBOA) –FFT part Conservatively, this requires 8 complex multiplies and 22.4 complex adds per clock cycle at 128MHz Looking at complex multiplies, we need 8 x 128 = 1024 MOPS –Frequency-domain EQ 100 data carriers multiplied by their respective conjugate channel taps every 312.5ns Implies 100/ = 320 MOPS Overall: 1344 M complex multiplies/sec MBOA OFDM has better performance and is computationally less expensive.