March, 2010 Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs) Submission Title: Integration lengths for long-range PHY Date Submitted: 16th March 2010 Source: Andy Ward, Ubisense Address: St Andrew’s House, St Andrew’s Road, Chesterton, Cambridge, CB4 1DL, ENGLAND Voice: +44 1223 535170, FAX: +44 1223 535167, E-Mail: andy.ward@ubisense.net Re: TG4f Call for Preliminary Proposals and Final Proposals, IEEE P802.15-09-0419-01-004f Abstract: Integration lengths for long-range PHY Purpose: To be considered by 802.15TG4f Notice: This document has been prepared to assist the IEEE P802.15. 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 P802.15. Andy Ward, Ubisense

Integration lengths for long-range PHY March, 2010 Integration lengths for long-range PHY Andy Ward Ubisense Andy Ward, Ubisense

doc.: IEEE 802.15-<doc#> <month year> doc.: IEEE 802.15-<doc#> March, 2010 Overview Discuss trade-offs involved in selecting the pulse-to-symbol ratio Show how system frequency accuracy is important Discuss ‘state-of-the-art’ in cheap crystal technology Andy Ward, Ubisense <author>, <company>

March, 2010 Long-range UWB PHY Rather than using one pulse-per-symbol mapping, we use m pulses-per-symbol Still very simple for transmitter to generate Still relatively simple for receivers as well! Use a different PRF (2MHz) so that receivers can do tone detection to figure out what type of packet is coming Integrate the pulses at the receiver to increase signal to noise For example, integration of four pulses increases SNR by 6dB (ideally) System will be average-power limited rather than peak-power limited Long pulse trains will mean that ‘average in 1ms’ trick can’t be used However, once you are average-power limited, there is no (regulatory) limit to the length of packet you can transmit at the same power Andy Ward, Ubisense

Ideal coherent integrator performance March, 2010 Ideal coherent integrator performance Andy Ward, Ubisense

System frequency accuracy March, 2010 System frequency accuracy Nominal pulse centre time The previous graph assumes perfect frequency sources throughout the system When the frequency sources differ, the pulses won’t line up perfectly in the coherent receiver So the achievable gain is less As the number of pulses-per-symbol increases, frequency accuracy gets more important You’re integrating over a longer time… …so the amount of drift between the transmitter and receiver increases… …so the pulses at the start and end of the integration period don’t line up so well Some time later…. Nominal pulse centre time Andy Ward, Ubisense

Effect of system frequency accuracy on integration gain March, 2010 Effect of system frequency accuracy on integration gain Andy Ward, Ubisense

System frequency accuracy vs centre frequency March, 2010 System frequency accuracy vs centre frequency Nominal pulse centre time There is some dependence of integration performance (with non-ideal frequency sources) on system centre frequency As CF increases, it takes less drift to make the pulses align less well in the coherent integrator So you need a higher frequency accuracy to maintain integration gain at high pulse-to-symbol numbers NB: What is shown here is a half-cycle of the CF of the UWB pulse (which will typically consist of a few complete cycles in a short-duration envelope) Some time later…. Nominal pulse centre time Extra energy for red (low frequency) signal Extra energy for blue (high frequency) signal Andy Ward, Ubisense

System frequency accuracy vs CF March, 2010 System frequency accuracy vs CF Andy Ward, Ubisense

How does integration gain vary against packet length? March, 2010 How does integration gain vary against packet length? Here, both systems are average-limited so graph flattens out Here, 1-pulse-per-symbol system is peak-limited, so graph flattens out NB: This graph is shows relative performance of the 2MHz 64-pulse-per-symbol and 1MHz 1-pulse-per-symbol modes. The increase in relative effective gain as packet length increases is due to the decrease in performance of the 1-pulse-per-symbol mode at packet lengths beyond 185 bits (up to 1000 bits). It does not indicate (nor is it intended to) an increase in absolute gain of the 64-pulse-per-symbol mode as packet length increases. Andy Ward, Ubisense

What is a reasonable frequency tolerance to aim for? March, 2010 What is a reasonable frequency tolerance to aim for? Remember – this is ONLY relevant to systems that want to do coherent integration. You don’t need a super crystal if you don’t care about this! Plain crystal – maybe 10ppm TCXO – maybe 0.3ppm 0.5ppm TCXOs over -40C to 85C are now readily available Used for GPS/GSM Very cheap Need to cope with other effects too: Initial frequency accuracy (tune out at manufacture) Crystal ageing (either tune out during use, or pre-age) As a data point (nothing more!), system frequency accuracy budget in current Ubisense systems is better than 2ppm Better than 1ppm would be achievable at same cost today Andy Ward, Ubisense

March, 2010 Proposal Based on available TCXOs and integration performance, we propose a 2MHz 64-pulse-to-symbol mapping for the long-range UWB PHY mode, and a 2.3ppm system timing accuracy This integration length and timing accuracy gives a relative gain over the 1MHz 1-pulse-per-symbol mapping of 4.5dB at 8.5GHz centre frequency and packet length of 185 bits This number increases as: Packet length increases (up to 1000 bits) Centre frequency decreases Timing accuracy increases Andy Ward, Ubisense