Massive testbed: Validation SEFFEN MALKOWSKY, LUND UNIVERSIY, SWEDEN
Why do we need a testbed? Ø heoretical results show that massive has very promising features for future wireless generation, however they cannot answer all questions: 1) How does it perform under real-life conditions? 2) How can we model massive channels? 3) Implementation challenges/issues? 4) Required reciprocity calibration accuracy? 5) Is pilot contamination a problem? 2
3 LUMAMI ESBED
LuMaMi estbed: Parameters Ø 100 BS antennas Ø Serving up to 12 UEs Ø 50 SDRs + 6 SDRs as UEs Ø 4 entral Processing units performing processing Ø 3.7 GHz carrier frequency Ø LE-like parameters Ø Uncoded transmission 4
LuMaMi estbed: Processing OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX Antenna ombiner/ BW splitter hannel Estimate Precoder Generate MMSE Matrix W MMSE Detector QAM Demapping QAM Mapping Up to (4) Processor FPGAs PXIe-8135 ontroller Group of 8 USRP RIOs OFDM X/RX OFDM X/RX OFDM X/RX hannel Estimate Precoder Generate MMSE Matrix WMM SE Detector QAM Demapping QAM Mapping NI USRP-2943R SDR PXIe-7976R FlexRIO OFDM X/RX Up to (8) Groups of 8 USRP RIOs (16 antennas per group) OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX BW ombiner/ Antenna Splitter Antenna om bi ner/ BW splitter BW ombiner/ Antenna Splitter hannel Estimate Precoder hannel Estimate Precoder Generate MMSE Matrix WMM SE Generate MMSE Matrix WMM SE Detector Detector QAM Demapping QAM Mapping QAM Demapping QAM Mapping BW ombining (bytes) System onfiguration BW Splitting (bytes) Video Display hannel Visualization Video Streaming 5 Ø Per-antenna (OFDM) processing performed on USRP SDR Ø Data of up to 8 USRPs combined and send to centralized processing
LuMaMi estbed: Processing OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX Antenna ombiner/ BW splitter hannel Estimate Precoder Generate MMSE Matrix W MMSE Detector QAM Demapping QAM Mapping Up to (4) Processor FPGAs PXIe-8135 ontroller Group of 8 USRP RIOs OFDM X/RX OFDM X/RX OFDM X/RX hannel Estimate Precoder Generate MMSE Matrix WMM SE Detector QAM Demapping QAM Mapping NI USRP-2943R SDR PXIe-7976R FlexRIO OFDM X/RX Up to (8) Groups of 8 USRP RIOs (16 antennas per group) OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX BW ombiner/ Antenna Splitter Antenna om bi ner/ BW splitter BW ombiner/ Antenna Splitter hannel Estimate Precoder hannel Estimate Precoder Generate MMSE Matrix WMM SE Generate MMSE Matrix WMM SE Detector Detector QAM Demapping QAM Mapping QAM Demapping QAM Mapping BW ombining (bytes) System onfiguration BW Splitting (bytes) Video Display hannel Visualization Video Streaming 6 Ø entralized processing distributed over 4 units, each working on ¼ bandwidth Ø 3 detector types implemented: MR/MR, ZF and regularized ZF
LuMaMi estbed: Processing OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX Antenna ombiner/ BW splitter hannel Estimate Precoder Generate MMSE Matrix W MMSE Detector QAM Demapping QAM Mapping Up to (4) Processor FPGAs PXIe-8135 ontroller Group of 8 USRP RIOs OFDM X/RX OFDM X/RX OFDM X/RX hannel Estimate Precoder Generate MMSE Matrix WMM SE Detector QAM Demapping QAM Mapping NI USRP-2943R SDR PXIe-7976R FlexRIO OFDM X/RX Up to (8) Groups of 8 USRP RIOs (16 antennas per group) OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX OFDM X/RX BW ombiner/ Antenna Splitter Antenna om bi ner/ BW splitter BW ombiner/ Antenna Splitter hannel Estimate Precoder hannel Estimate Precoder Generate MMSE Matrix WMM SE Generate MMSE Matrix WMM SE Detector Detector QAM Demapping QAM Mapping QAM Demapping QAM Mapping BW ombining (bytes) System onfiguration BW Splitting (bytes) Video Display hannel Visualization Video Streaming 7 Ø A standard windows computer hosts the complete system Ø onfigures system Ø Visualizes parameters, e.g. channel impulse response and acts as source and sink for video streams
8 FOUR ESS
Indoor Measurements Ø Serving 12 users in groups of 4 on UL and DL in a lecture hall Ø Record BERs while sweeping amplifier gains (0-30 db) Ø UL: Sweep the gain of the UEs transmitter Ø DL: Sweep the gain of the BS transmitters Ø 18 Million bits transmitted per step Ø ompare performance of MR/MR vs. ZF 9
Setup BS Base station placed at the front of the lecture hall losely spaced UEs Users are distributed in groups of 4 with close spacing inside each group 10
Uplink BERs (QPSK) 11 BER 10 0 10-1 10-2 10-3 10-4 0 1 2 3 UE0 UE1 UE2 UE3 UE4 UE5 UE6 UE7 UE8 UE9 UE10 UE11 4 5 6 7 0 10 20 30 Amplifier Gain [db] 9 10 11 X saturates 8 BS Ø ZF detector Ø lose by users (UE8-11) show best BERs Ø Far users (UE0-UE3) show worst performance Ø UE0/1 interference limited Ø Bad performance at full power close to saturation
Uplink BERs (16-QAM) BER 12 10 0 10-1 10-2 10-3 10-4 0 1 2 3 UE0 UE1 UE2 UE3 UE4 UE5 UE6 UE7 UE8 UE9 UE10 UE11 0 10 20 30 Gain Offset [db] 4 5 6 7 9 10 11 8 BS X saturates Ø ZF detector Ø lose by users (UE8-11) shows best BERs Ø Far users (UE0-UE3) show worst performance Ø UE0/1 interference limited Ø Bad performance at full power close to saturation
Downlink BERs (QPSK) 13 BER 10 0 10-1 10-2 10-3 10-4 0 1 2 3 UE0 UE1 UE2 UE3 UE4 UE5 UE6 UE7 UE8 UE9 UE10 UE11 0 5 10 15 Amplifier Gain [db] 4 5 6 7 9 10 11 8 BS Ø ZF precoder Ø lose by users (UE8-11) shows best BERs Ø Other users (UE0-7) show similar performance Ø High performance difference between UE8-11 and the UE0-7
Downlink BERs (16-QAM) BER 14 10 0 10-1 10-2 10-3 10-4 0 1 2 3 UE0 UE1 UE2 UE3 UE4 UE5 UE6 UE7 UE8 UE9 UE10 UE11 0 10 20 30 Gain Offset [db] 4 5 6 7 9 10 11 8 BS Drop in BER for closest users BERs floor Ø ZF detector Ø lose by users (UE8-11) show best BERs Ø Almost all users show a significant BER floor Ø Performance limited by reciprocity calibration accuracy and/or interference
ZF vs. MR/MR 10 0 Uplink 10 0 Downlink BER 10-1 10-2 BER 10-1 10-2 Floor for ZF with 64-QAM 10-3 10-4 MR QPSK MR 16-QAM MR 64-QAM ZF QPSK ZF 16-QAM ZF 64-QAM 0 10 20 30 Amplifier Gain [db] 10-3 10-4 MR QPSK MR 16-QAM MR 64-QAM ZF QPSK ZF 16-QAM ZF 64-QAM Ø MR/MR showing significant error floors Ø Usable in practical systems? Ø ZF far superior in real-life channels 0 10 20 30 Amplifier Gain [db] 15
ollaborative Measurements in Bristol Ø First joint measurement campaign with Lund University and University of Bristol in May 2016 Ø Based on Bristol MaMi system with 128 antennas Ø How many users can be served simultaneously? Ø What spectral efficiency can be achieved? 16
Setup Base station losely spaced users Users BS 17 Overall setup from base station view
Results Ø 22 users served simultaneously using 256-QAM Ø Uncoded transmission Ø Equating a spectral efficiency of 145 b/s/hz World record spectral efficiency 18
First Outdoor rial I Ø First outdoor test to see whether reciprocity calibration and over-the-air synchronization works outdoor Ø Focus on achieving good UL and DL constellations with up to 8 users and an additional moving user UE UE BS UE 19
First Outdoor rial III ZF UL constellations on BS MR DL constellations with ZF 20 lo d n i w se lf a h ow
First Outdoor rial III One user in car Front Panel picture g n i k o o l d Goo ellation t s n o c 21
Joint Mobility rials in Lund Ø ogether with a team from University of Bristol we performed the first mobility test using the LuMaMi testbed Ø Goal was to analyze how well massive works in dynamic environments Ø ests were performed with up to 12 users, some mounted on cars (up to 40 km/h) and some mounted on cycle carts (walking speed) Ø Uplink channel data, BERs and LabVIEW front panels were recorded 22
Setup / Equipment Base station deployed on rooftop Users mounted on cycle cart 23 Users mounted on cars
Different tested scenarios I AR art art art art AR Ø Gradually adding users while observing performance Parking lot 24
Different tested scenarios II Parking lot AR art art art AR Ø Scenario with 6 users as pedestrians and 4 users on cars Ø Video including live BERs and constellations Ø Playback recorded channel data 25
10 user mobility test video PLEASE HEK OU FOLLOWING LINK https://www.youtube.com/watch?v=wppmrr4rhmo 26
onclusions Ø he LuMaMi testbed is fully functional and working for UL and DL transmission Ø In a joint campaign Lund University and University of Bristol achieved a new world record for spectral efficiency Ø Indoor measurements showed that even with 100 antennas at the BS, MR/MR show significantly worse performance than ZF Ø Mobility measurements showed that massive works for moving users with relatively good BER performance on UL and DL (more analysis to be performed) 27
Frame Structure One radio frame f = 10ms Subframe #0 Subframe #1 Subframe #2 Subframe #3 Subframe #4 Subframe #9 One subframe sf = 1ms One slot slot = 0.5ms ULPilot ULData ULData Switch Guard DLData DLData Switch Guard 28