Downlink Multi-User MIMO for IEEE 802.16m Sivakishore Reddy Naga - PowerPoint PPT Presentation

Downlink Multi-User MIMO for IEEE 802.16m Sivakishore Reddy Naga Sekhar Centre of Excellence in Wireless Technology 2013

Outline 1 Introduction 2 Closed Loop MU-MIMO 3 Results 4 Open Loop MU-MIMO 5 Results 6 Conclusions Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 2 / 40

Outline 1 Introduction 2 Closed Loop MU-MIMO 3 Results 4 Open Loop MU-MIMO 5 Results 6 Conclusions Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 3 / 40

MIMO Introduction Figure: MIMO Systems Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 4 / 40

SU-MIMO vs MU-MIMO Single User MIMO : Promises reliability and channel capacity through diversity gain and rate maximization. Mainly acts as physical(PHY) layer performance booster. Multi User MIMO : Uses spatial degrees of freedom to schedule multiple users to simultaneously share the spatial channel. Optimum design strategy to increase the system capacity. Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 5 / 40

Requirements of MU-MIMO : Requires Channel State Information. Partial CSI. Codebook based precoding. Requires more complex scheduling algorithms. Requires advanced transceiver methodologies. Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 6 / 40

Outline 1 Introduction 2 Closed Loop MU-MIMO 3 Results 4 Open Loop MU-MIMO 5 Results 6 Conclusions Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 7 / 40

System model 1 ^ x1 . . User 1 . N User1 data r x 1 1 . PRECODER (W) PF SCHEDULER x User 2 data 2 . . 2 . . 1 . . . ^ x 2 . . User 2 x K . User N data u N r N t . . 1 ^ . x K . User K . N r . . . Feedback: CQI, PMI Figure: MU-MIMO System model Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 8 / 40

Notation N t : No.of antennas at the BS. N r : No.of antennas at the MS/user. N u : No.of users contending for resource unit. K : No.of users served per resource unit. H : N r × N t desired channel matrix. G : N r × N t interfering co-channel matrix. W : N t × K precoder. v : N t × 1 precoding vector chosen from codebook. I : Number of strong interferers. Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 9 / 40

About Codebook based precoding In 16m Codebook is defined for a given MIMO setup. � � W = v 1 v 2 . . . v K N t × K The N r × 1 received signal vector at the k th MS : y k = H k Wx + n k , k ∈ { 1 , 2 , . . . , K } K � � P P � = K H k v k x k + H k v i x i + n k K i � = k,i =1 Desired signal IUI Noise � T � � � � P P P x = K x 1 K x 2 . . . K x K K × 1 Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 10 / 40

CL-MUMIMO PROBLEM STATEMENT How should each user make his choice for the PMI ? How should each user model his CQI ? How should the BS use this feedback-info to schedule multiple users per data region? Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 11 / 40

Choose PMI to minimize MSE MUMIMO Equation : I � G i W ′ y k = H k Wx + i z i + n k i =1 MSE for i th precoding vector is : x s ) 2 = E ( x s − ˆ MSE i � − 1 H k v i x s − v ∗ i H ∗ H k WW ∗ H ∗ σ 2 � k + I cov + σ 2 I = k I � G i W ′ i W ′∗ i G ∗ I cov = i i =1 If W is unitary, � − 1 H k v i σ 2 x s − v ∗ i H ∗ H k H ∗ k + I cov + σ 2 I � MSE i = k � � � PMI = arg min MSE i i ∀ subcarriers Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 12 / 40

CQI modeling Receiver uses an MMSE filter b , K I � � b ∗ b ∗ b ∗ b ∗ k G i W ′ i z i + b ∗ k y k = k H k v k x k + k H k v i x i + k n k i � = k,i =1 i =1 | b ∗ k H k v k | 2 CQI = � K k H k v k | 2 + b ∗ i � = k,i =1 | b ∗ k ( I cov + σ 2 I ) b k where I cov = � I i = � I i =1 G i W ′ i W ′∗ i G ∗ i =1 G i G ∗ i Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 13 / 40

MUMIMO PF-Scheduler Find all possible pairs of users who have reported orthogonal PMI. Find sum-PF-metric for each pair. Schedule pair with maximum sum-PF-metric. Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 14 / 40

Outline 1 Introduction 2 Closed Loop MU-MIMO 3 Results 4 Open Loop MU-MIMO 5 Results 6 Conclusions Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 15 / 40

Unitary vs Non-Unitary Precoders The covariance term which influence CQI : I � G i W ′ i W ′∗ i G ∗ I cov = i i =1 If the precoder W is not unitary, the interference level may change from one frame to another frame as the precoders used changes. So the CQI of 2x2 is stable compared to 4x2 as the precoders of 4x2 system are non-unitary. Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 16 / 40

CQI Stability Comparision of Histograms 1 2x2 4x2 0.8 Histogram 0.6 0.4 0.2 0 −10 −5 0 5 10 15 CQI(n−1)−PPSINR(n) in dB Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 17 / 40

2x2 vs 4x2 CDF of User Throughputs 1 0.8 2x2 Ideal I cov 4x2 Ideal I cov 0.6 CDF 0.4 0.2 0 0 500 1000 1500 2000 2500 3000 3500 4000 4500 User Throughput (in kbps) Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 18 / 40

Practical Implications in Closed loop system The mimo modes of interfering users can change in closed loop region. The calculation of CQI which involves I cov cannot be perfectly determined. The PMI decision term also involves I cov is also not perfect. In demodulation, if the mimo mode of CCI is different it is difficult to even estimate the interference covariance at the reciever. Thus degrading the performance further. The PMI and CQI fedback are IMPERFECT and reduces the capacity. Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 19 / 40

CL-MUMIMO CDF of User Throughputs 1 0.8 2x2 Ideal I cov Practical 4x2 2x2 Scaled I cov 0.6 4x2 Ideal I cov CDF 4x2 Scaled I cov 0.4 Practical 2x2 0.2 0 0 500 1000 1500 2000 2500 3000 3500 4000 4500 User Throughput (in kbps) Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 20 / 40

CL-MU-MIMO vs OL-SU-MIMO 3 OL−SU−MIMO,2x2 2.8 CL−MU−MIMO,2x2 Sector Spectral Efficiency 2.6 in bps/Hz 2.4 2.2 2 1.8 5 10 15 20 25 30 35 40 45 50 #Active users per sector Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 21 / 40

Outline 1 Introduction 2 Closed Loop MU-MIMO 3 Results 4 Open Loop MU-MIMO 5 Results 6 Conclusions Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 22 / 40

Open Loop MU-MIMO System Model 1 ^ x1 . . User 1 . N User1 data r x 1 1 . PRECODER (W) PF SCHEDULER x User 2 data 2 . . 2 . . 1 . . ^ . x 2 . . User 2 x K . User N data u N r N t . . 1 . ^ x K . User K . N r . . . Feedback: CQI, PSI Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 23 / 40

Open Loop Multi-User MIMO Precoders are fixed a priori. � � W = v 1 v 2 . . . v K N t × K Each user feedbacks preferred stream index ( PSI ) and CQI . PSI ∈ { 1 , 2 , . . . , K } . PF scheduler will serve set of K users who feedback PSIs { 1 , 2 , . . . , K } . Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 24 / 40

Open Loop Region Inside Open Loop Region : All the base stations will use same MIMO mode. Creates stable interference environment. Estimation of CCI is easy and accurate. Covariance matrix is calculated using estimated precoded channels. Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 25 / 40

Received signal The N r × 1 received signal vector at the k th MS : 8 � G ik Wx ′ y k = H k Wx + ik + n k , k ∈ { 1 , 2 , . . . , K } i =1 K 8 � � P P � � G ik Wx ′ = K H k v k x k + H k v i x i + ik + n k K i =1 i � = k,i =1 Desired signal IUI CCI Noise � T � � � � P P P x = K x 1 K x 2 . . . K x K K × 1 Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 26 / 40

Covariance matrix and MMSE filter The Covariance matrix of CCI ( K CCI ) : � P 8 � � G ik WW ∗ G ∗ K CCI = ik K i =1 1 × N r MMSE filter equation for k th user : � − 1 � k + K P K CCI + KN o b k = ( H k v k ) ∗ H k ˜ ˜ H ∗ I N r P where ˜ H k = H k W . Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 27 / 40

SINR Calculation x k = b k y k ˆ K 8 � � P P � � b k G ik Wx ′ = K b k H k v k x k + b k H k v i x i + ik + b k n k K i � = k,i =1 i =1 � P � | b k H k v k | 2 K SINR k = IUI k + CCI k + � b k � 2 N o � P K � � | b k H k v i | 2 IUI k = K i � = k,i =1 CCI k = b k K CCI b ∗ k,l Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 28 / 40

CQI and PSI Computation � � W = v 1 v 2 . . . v K N t × K � P � | b k H k v l | 2 K SINR k,l = IUI l + CCI k + � b k � 2 N o PSI k = arg max SINR k,l l CQI k = SINR k,P SI k Sivakishore Reddy Naga Sekhar (CEWiT) MU-MIMO for IEEE 802.16m 2013 29 / 40