Real Interference Alignment for the MIMO Multiple Access Wiretap Channel
Pritam Mukherjee S¸ennur Uluku¸s
University of Maryland, College Park
The MIMO Multiple Access Wiretap Channel (MAC-WT)
I Consider the two-user N × N × N × K MIMO MAC-WT:
H1
G2
H2 G1
N antennas N antennas
N antennas K antennas W1
W2
Wˆ1, ˆW2
W1W2
I The channel gains arefixed across time-slots.
I All channel gains are known perfectlyat every terminal.
I Question: What is the optimalsum secure degrees of freedom?
Prior Work: The SISO MAC-WT Channel
I For the case N = K = 1, i.e., the SISO MAC-WT channel:
h1
g2
h2 g1
W1
W2
Wˆ1, ˆW2
W1W2 I Optimal sum s.d.o.f.a= 23.
I Achievable scheme based onreal interference alignment.
Prior Work: The SISO MAC-WT Channel (contd.)
I The alignment of signals in this case is as follows:
h1
g2
g1
h2
v1
v2
u1
u2
u1
v1 v2
v2
u2
v1
u1
u2
Prior Work: The Fading MIMO MAC-WT Channel
I Thefadingtwo-user N × N × N × K MIMO MAC-WT:
H1(t)
G2(t) H2(t) G1(t)
N antennas N antennas
N antennas K antennas W1
W2
Wˆ1, ˆW2
W1W2
I The channel gains are i.i.d. across time slots.
I Question: What is the optimalsum secure degrees of freedom?
The Fading MIMO MAC-WT Channel (contd.)
I Theorem: [Mukherjee, Ulukus, Asilomar 2015]: The optimalsum s.d.o.f. of the N × N × N × K fading MIMO MAC-WT is
ds =
N, if K ≤ 12N
2
3(2N − K ), if 12N ≤ K ≤ N
2
3N, if N ≤ K ≤43N 2N − K , if 43N ≤ K ≤ 2N
0, if K ≥ 2N.
I Note that when N = K = 1, the optimalsum s.d.o.f.a=23.
a[Xie, Ulukus, 2013]
The Fading MIMO MAC-WT Channel (contd.)
N 2N/3
N2 4N
N 3 2N K
sum s.d.o.f.
multiple access wire- tap channel
I Converseproof holds forfixed channel gains as well.
I Question: Is the same s.d.o.f. achievable withfixedchannel gains? Yes!
The Fading MIMO MAC-WT Channel (contd.)
N 2N/3
N2 4N
N 3 2N K
sum s.d.o.f.
N/2
3N2
wiretap channel with one helper
multiple access wire- tap channel
I Converseproof holds forfixed channel gains as well.
I Question: Is the same s.d.o.f. achievable withfixedchannel gains? Yes!
The Fading MIMO MAC-WT Channel (contd.)
N 2N/3
N2 4N
N 3 2N K
sum s.d.o.f.
N/2
3N2
wiretap channel with one helper
multiple access wire- tap channel
I Converseproof holds forfixed channel gains as well.
I Question: Is the same s.d.o.f. achievable withfixedchannel gains?
Yes!
The Fading MIMO MAC-WT Channel (contd.)
N 2N/3
N2 4N
N 3 2N K
sum s.d.o.f.
N/2
3N2
wiretap channel with one helper
multiple access wire- tap channel
I Converseproof holds forfixed channel gains as well.
I Question: Is the same s.d.o.f. achievable withfixedchannel gains?
Yes!
Achievable Scheme for K ≤
N2N 2N /3
N2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: K ≤ 3.
I Optimal sum s.d.o.f.=N.
I Beam-formingis optimal when K ≤ N2.
I Transmitters 1 and 2 sendN − K andK symbols v1and v2, respectively, in the nullspace of the eavesdropper’s channels.
I Note that K ≤ N − K in this regime.
Achievable Scheme for K ≤
N2N 2N /3
N2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: K ≤ 3.
I Optimal sum s.d.o.f.=N.
I Beam-formingis optimal when K ≤ N2.
I Transmitters 1 and 2 sendN − K andK symbols v1and v2, respectively, in the nullspace of the eavesdropper’s channels.
I Note that K ≤ N − K in this regime.
Achievable Scheme for
4N3≤ K ≤
3N2N 2N /3
N
2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
Achievable Scheme for
4N3≤ K ≤
3N2N 2N /3
N
2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: 8 ≤ K ≤ 9.
I Optimal sum s.d.o.f.= 2N − K .
I Transmitter 1:
v1∈ R3N−2K, ˜v ∈ R3K −4N
, i.e.,K − N symbols.
I Transmitter 2:
v2∈ R3N−2K
, i.e.,3N − 2K symbols.
I Total of2N − K symbols in each time slot.
Achievable Scheme for
4N3≤ K ≤
3N2(contd.)
I The channel inputs are:
X1= R1˜v+ P1v1+ H−11 Qu1 X2= R2u˜+ P2v2+ H−12 Qu2
I The channel outputs are:
Y =H1R1˜v+ H1P1v1+ H2P2v2+ H2R2˜u+ Q(u1+u2)
Z =G1R1˜v+ G2R2˜u+ G1P1v1+ G2H−12 Qu2+ G2P2v2+ G1H−11 Qu1
I For security, enforce:
G1R1=G2R2 G1P1=G2H−12 Q G2P2=G1H−11 Q
Achievable Scheme for
4N3≤ K ≤
3N2(contd.)
I Feasibility of G1R1= G2R2with N × (3K − 4N) matrix Ri: [G1 −G2]
R1 R2
= 0
I This is feasible since 3K − 4N ≤ 2N − K in this regime.
I Choose Pi and Q as solutions of:
G1 0K ×N −G2H−12 0K ×N G2 −G1H−11
P1 P2 Q
= 0
I Security: Guaranteed by design.
I Decodability: Number of symbols to decode:
(2N − K )
| {z }
desired symbols
+ (3K − 4N)
| {z }
˜ u
+ (3N − 2K )
| {z }
u1+u2
= N
Achievable Scheme for
4N3≤ K ≤
3N2(contd.)
I The alignment structure in this case has the following form:
H1
G2 G1
H2
˜
v v1
v2 u1
u2
u1
v2 u2
v1
˜
u ˜u
˜ v
˜
v v1 v2 ˜u
u1
u2
Achievable Scheme for
3N2≤ K ≤ 2N
N 2N /3
N2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: 9 ≤ K ≤ 12.
I Optimal sum s.d.o.f.= 2N − K .
I TheMAC-WTreduces to thewiretap channel with one helpera.
I Transmitter 1: v ∈ R2N−K, i.e.,(2N − K )information symbols.
I Transmitter 2 sends only cooperative jammingsignals.
a[Nafea, Yener, 2015]
Achievable Scheme for
3N2≤ K ≤ 2N
N 2N /3
N2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: 9 ≤ K ≤ 12.
I Optimal sum s.d.o.f.= 2N − K .
I TheMAC-WTreduces to thewiretap channel with one helpera.
I Transmitter 1: v ∈ R2N−K, i.e.,(2N − K )information symbols.
I Transmitter 2 sends only cooperative jammingsignals.
Achievable Scheme for
3N2≤ K ≤ 2N (contd.)
I The channel inputs are:
X1= Pv X2= Qu
I The received signals are
Y =H1Pv+ H2Qu Z =G1Pv+ G2Qu
I For security, choose P and Q as the solutions to [G1 −G2]
P Q
= 0
I Decodability: Receiver can decode both v and u, since 2(2N − K ) ≤ N
Achievable Scheme for
3N2≤ K ≤ 2N (contd.)
I The alignment structure in this case has the following form:
u u
v u v v
H1
G2
G1
H2
Achievable Scheme for
N2≤ K ≤ N
N 2N /3
N
2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: 3 ≤ K ≤ 6.
I Optimal sum s.d.o.f.=23(2N − K ) = 2 N − K + d +3l , where d =
2K − N 3
, l = (2N − K ) mod 3
I Examples:
1. When K = 4, sum s.d.o.f.=163, d = 0, l = 2. 2. When K = 5, sum s.d.o.f.=143, d = 1, l = 1.
Achievable Scheme for
N2≤ K ≤ N
N 2N /3
N
2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: 3 ≤ K ≤ 6.
I Optimal sum s.d.o.f.=23(2N − K ) = 2 N − K + d +3l , where d =
2K − N 3
, l = (2N − K ) mod 3
I Examples:
16
Achievable Scheme for
N2≤ K ≤ N (contd.)
I Main idea: Decompose the channel input at each transmitter into:
1. N − K Gaussiansymbols sent in the nullspace of Eve.
2. Gaussiansymbols carryingd s.d.o.f.
3. structured PAMsymbols carrying 3l s.d.o.f.
I Use channel precoding for the Gaussiansymbols.
I Use real alignment schemes for thel × l × l × l MAC-WT achieving
2l
3 sum s.d.o.f. for l = 1, 2.
I For l = 1 : Real alignment scheme for the SISO MAC-WT is knowna.
I Needed: Real alignment scheme for the 2 × 2 × 2 × 2 MAC-WT.
a[Xie, Ulukus, 2013]
Achievable Scheme for the 2 × 2 × 2 × 2 MAC-WT channel
I Optimal sum s.d.o.f.=43.
H1
G2
G1
H2
v11 v12
v21 v22
v12 v22
v11 v21
u11 u12
u21 u22
u11
u21
v11
v21
v12
v22
u12
u22
u11
v21
u12
v22
u21
v11
u22
v12
asymptotic alignment
) perfect alignment
) perfect alignment
)
asymptotic alignment )
I Perfect alignment at the eavesdropper ensures security.
I At receiver, d.o.f. at each antenna = 23; total s.d.o.f. =43.
The General Scheme for
N2≤ K ≤ N
I Each transmitter wants to send N − K + d + 3l s.d.o.f.
I At transmitter i, information bearing symbols: (˜vi,v(1)i ,v(2)i )
I ˜vi :N − K Gaussian symbols that can be sent using Eve’s nullspace
I v(1)i : d Gaussiansymbols each carrying 1 d.o.f.
I v(2)i :l structuredsymbols each carrying 13 d.o.f.
I Cooperative jamming signals: ui= (u(1)i ,u(2)i ).
I Let vi = (v(1)i ,v(2)i ). Transmitter i sends:
Xi =G⊥i v˜i+ Pivi+ H−1i Qui
I The received signals are:
Y =H1G⊥1˜v1+ H1P1v1+ H2P2v2+ H2G⊥2˜v2+ Q(u1+u2) + N1
Z =G1P1v1+ G2H−12 Qu2+ G2P2v2+ G1H−11 Qu1+ N2
The General Scheme for
N2≤ K ≤ N (contd.)
I Let Q to be any N × (d + l) matrix with full column rank.
I Set Pi = GTi (GiGTi )−1(GjH−1j )Q.
I Eve’s observation is:
Z = G2H−12 Q(v1+u2) + G1H−11 Q(v2+u1) + N2
I Perfect alignment ensures security.
I Decoding: Consider BN×l such that
BT[H1G⊥1 H2G⊥2 H1P(1)1 H1P(1)1 Q(1)]N×(N−l)= 0
I Consider ˜Y = (BTQ(2))−1BTY
Y = DH˜ 1P(2)1 v(2)1 + DH2P(2)2 v(2)2 + (u(2)1 +u(2)2 ) + DN1
I Decode (v(2)1 ,v2(2)) using the l × l × l × l MAC-WT scheme.
(2) (2) (2) (2) (1) (1)
Achievable Scheme for N ≤ K ≤
4N3N 2N /3
N
2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: 6 ≤ K ≤ 8.
I Optimal sum s.d.o.f.=2N3 .
I The point K = N is achievable using the scheme for N2 ≤K ≤ N.
I The point K = 4N3 is achievable using the scheme for 4N3 ≤K ≤ 3N2 .
I The intermediate points N ≤ K ≤ 4N3 are achievable since increasing Eve’s antennas does not increase the sum s.d.o.f.
Achievable Scheme for N ≤ K ≤
4N3N 2N /3
N
2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: 6 ≤ K ≤ 8.
I Optimal sum s.d.o.f.=2N3 .
I The point K = N is achievable using the scheme for N2 ≤K ≤ N.
I The point K = 4N3 is achievable using the scheme for 4N3 ≤K ≤ 3N2 .
I The intermediate points N ≤ K ≤ 4N3 are achievable since increasing
Achievable Scheme for K ≥ 2N
N 2N /3
N2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: K ≥ 12.
I Optimal sum s.d.o.f.=0.
I Since Eve has more than 2N antennas, the input of both transmitters can be decoded to within noise variance.
Achievable Scheme for K ≥ 2N
N 2N /3
N2 4N
N 3 2N K
s.d.o.f.
(0, 0) 3N
2
I Running example: N = 6. This case: K ≥ 12.
I Optimal sum s.d.o.f.=0.
I Since Eve has more than 2N antennas, the input of both transmitters can be decoded to within noise variance.
Conclusions and Future Work
I Provided achievable schemes for the MIMO MAC-WT with fixed channel gains.
I The achievable scheme for the regime N2 < K < N:
I Is based onasymptoticreal interference alignment
I Uses a combination ofGaussianandstructured PAMsymbols.
I Combines channel precoding with real interference alignment
I Open question: What happens if Eve’s CSIT is not available?