ebook img

Generalization of Pigeon Hole Bound PDF

0.05 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Generalization of Pigeon Hole Bound

1 Generalization of Pigeon Hole Bound Toni Ernvall Department of Mathematics, University of Turku Abstract—H.-F. Lu, J. Lahtonen, R. Vehkalahti, and C. Hol- Then we have det(AA†)=det(BB†). lanti introducedso called Pigeon Hole Boundfor decayfunction Proof: If k > n then det(AA†) = 0 = det(BB†). If 2 of MIMO-MAC codes. Here we give a generalization for it. k =n then det(A) is 1 c e e e 20 I. DECAY FUNCTION (cid:12)(cid:12) c12 (cid:12)(cid:12) (cid:12)(cid:12) c21 (cid:12)(cid:12) (cid:12)(cid:12) e12 (cid:12)(cid:12) (cid:12)(cid:12) e12 (cid:12)(cid:12) WeconsiderheredecayfunctionofMIMO-MACcodes.Ev- (cid:12) . (cid:12) (cid:12) . (cid:12) (cid:12) . (cid:12) (cid:12) . (cid:12) n ery user has some specific lattice L ⊆ M ,j = 1,...,U (cid:12)(cid:12) .. (cid:12)(cid:12)=(cid:12)(cid:12) .. (cid:12)(cid:12)=(cid:12)(cid:12) .. (cid:12)(cid:12)=···=(cid:12)(cid:12) .. (cid:12)(cid:12) a j n×k (cid:12) c (cid:12) (cid:12) c (cid:12) (cid:12) c (cid:12) (cid:12) e (cid:12) J withk ≥Un.Weassumethateachuser’slatticeisoffullrank (cid:12) k−1 (cid:12) (cid:12) k−1 (cid:12) (cid:12) k−1 (cid:12) (cid:12) k−1 (cid:12) 7 r =2kn so the lattice Lj hasan integralbasis Bj,1,...,Bj,r. (cid:12)(cid:12)(cid:12) ck (cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12) ck (cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12) ck (cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12) ck (cid:12)(cid:12)(cid:12) 2 Nowthecodeassociatedwithjthuserisarestrictionoflattice i.e. d(cid:12)et(B) a(cid:12)nd h(cid:12)ence de(cid:12)t(AA(cid:12)†)=de(cid:12)t(BB†). (cid:12) (cid:12) Lj such that Assume k <n. Let v ,...,v ∈Cn be such that v ∈ ] 1 n−k 1 A r L(c ,c ,...,c ,)⊥\{0},v ∈L(v ,c ,c ,...,c )⊥\{0}, 1 2 k 2 1 1 2 k R Lj(Nj)={XbiBj,i|bi ∈Z,−Nj ≤bi ≤Nj} ...,vn−k ∈L(v1,v2,...vn−k−1,c1,c2,...,ck)⊥\{0}.Now . i=1 (as in the case n=k) we have h where N is a given positive number. j c e at Using these definitions the U-user mimo-mac code is  .1   .1  m (L (N ),L (N ),...,L (N )). .. .. 1 1 2 2 U U [ For this we define  ck−1   ek−1  1 D(N1,...,NU)= min det(MM†) det( vck )=det( vck ) v Xj∈Lj(Nj)\{0}  1   1      4 where M =M(X1,...,XU). For a special case N1 =···=  ...   ...  74 NU =N we write  vn−k   vn−k  5 D(N)=D(N1 =N,...,NU =N). and hence . 01 In the special case k =Un we have (cid:12) c1.c∗1 ... c1.c∗k c1.v1∗ ... c1v.n∗−k (cid:12) 2 D(N1,...,NU)=D(N1,...,NU)2 (cid:12)(cid:12) .. .. .. .. (cid:12)(cid:12) v:1 and especially D(N)=D(N)2. (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) vck1cc∗1∗1 ...... vck1cc∗k∗k vck1vv1∗1∗ ...... vck1vvn∗n∗−−kk (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Xi (cid:12)(cid:12) ... ... ... ... (cid:12)(cid:12) ar SIIo.cAalNledUPPPigEeRonBOHUoNleDBUoSuInNdGfoPIrGdEeOcaNyHfuOnLcEtiPoRnIoNfCMIPILMEO- (cid:12)(cid:12)(cid:12) vn−kc∗1 ... vn−kc∗k vn−kv1∗ ... vn−kvn∗−k (cid:12)(cid:12)(cid:12) (cid:12) (cid:12) is equal than MACcodeswasintroducedin[1].Here wegiveageneraliza- actiliosLno−efmeoimr ∈iat.2L.1(c:i+L1e,tcci1+,2c,2.,....,.c,kc)kc,1feo1r,ie2=,.1.,..,.e.k,−k1−∈1C.nW,arintde (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) vcek11...eee∗1∗1∗1 ......... vcek11...ccc∗k∗k∗k vcek11...vvv1∗1∗1∗ ......... vcek11vvv...nn∗∗n∗−−−kkk (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12).  c  (cid:12) .. .. .. .. (cid:12) 2 (cid:12) . . . . (cid:12) . (cid:12) (cid:12) A= ..  (cid:12)(cid:12) vn−kc∗1 ... vn−kc∗k vn−kv1∗ ... vn−kvn∗−k (cid:12)(cid:12)  ckc−1  A(cid:12)nd since the way we chose v1,...,vn−k this means tha(cid:12)t  k  c c∗ ... c c∗ 0 ... 0 and (cid:12) 1. 1 1. k . . (cid:12) e1 (cid:12)(cid:12) .. .. .. .. (cid:12)(cid:12) B = ecek...−k21 . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ck00...c∗1 ......... ck00...c∗k v100...v1∗ ......... vn−k00...vn∗−k (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12) (cid:12) 2 is equal than and especially K e e∗ ... e c∗ 0 ... 0 D(N)≤ (cid:12) 1. 1 1. k . . (cid:12) Nα (cid:12) . . . . (cid:12) (cid:12) . . . . (cid:12) where α= U−1 2n2(U−l). Especially if k =Un we have (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ck0...e∗1 ...... ck0...c∗k v10...v1∗ ...... 00... (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) PD(l=N11,k.−..n,(NU−Ul)) ≤KUY−1Nl−2n(Ul−l) (cid:12) (cid:12) l=1 (cid:12)(cid:12) 0 ... 0 0 ... vn−kvn∗−k (cid:12)(cid:12) and (cid:12) (cid:12) K because if we write e =c −x where x ∈L(c ,...,c ) D(N)≤ then v e∗ =v (c −xi )∗ =i v ci∗−v x∗i=0−0i+=1 0 for akll Nβ j i j i i j i j i i=1,...,n−1 and j =1,...,n−k. This gives that where β = U−1 2n(U−l). Proof:PLelt=u1s uselthe notation C =(c⊤ ,...,c⊤ )⊤ for |v |2...|v |2det(AA†)=|v |2...|v |2det(BB†) l l,1 l,n 1 n−k 1 n−k l=1,...,U. and hence det(AA†)=det(BB†). Let us first fix some small CU ∈ LU(NU). Now Lemma 2.2: Let V = Rn be an n-dimensional vector |CU| = O(1). Then write WU = {(x⊤1,...,x⊤n)⊤|xi ∈ space, N a given positive integer, and let c1,c2,...,cn ∈V cLo(mcUp,l1e,m..e.n,tcUan,nd)}π. Th:enMlet VU(C=) W→U⊥Vbe aitns oorrtthhooggoonnaall be a basis for V. Let also U be a k-dimensional subspace of U n×k U projection. V and π : V → U an orthogonal projection into U. We can choosea basis{π(c ),π(c ),...,π(c )}forU suchthatif A subspace VU has dimR(VU) = 2nk − dimR(WU) = v = a c +a c +j1···+aj2c with |ajk| ≤ N for all i then 2nk − 2n2 = 2n(k − n) so the image πU(LU−1(NU−1)) 1 1 2 2 n n i π(v)=b π(c )+b π(c )+···+b π(c )with|b |≤n2N fallsinto a 2n(n−k)-dimensionalhypercubewith side length 1 j1 2 j2 k jk i smaller or equal than (2nk)2N = O(N ) by lemma for all i. U−1 U−1 Proof: Since {c ,c ,...,c } is a basis for V we can 2.2 with coordinates having restricted length since projection choose a basis for U 1from2 the sent {π(c1),π(c2),...,π(cn)}. can only shrink. We also have |LU−1(NU−1)|=θ(NU2n−k1) so using the linearity of π and pigeon hole principle we have Without loss of generality we may assume that U π(c ),π(c ),...,π(c ) are linearly independent. We some CU−1 ∈LU−1(NU−1) such that 1 2 k say that they form a basis K . Letπ(ck+1)=d1π(c1)+d12π(c2)+···+dkπ(ck)forsome π (C )=O(2n(k−nvu) NU2n−(1k−n))=O(N−k−nn). d1,d2,...,dk ∈R. U U−1 ut NU2n−k1 U−1 If |d |≤1 for all i=1,2,...,k then let K =K . i 2 1 Now similarly build V = W⊥ for l = 0,...,U −2 πdk(+cO1jt)πhe(cr=wis−ed)dj+1leπ−t(dc|jd1+j)1|π+(>c−dd1j2)πb+(ec·2·a)·+m+−adx·k·im+·1+aπl(cc−odde)jj−fafi1ncπdie(tcnhjte.−a1Nb)soo+w- Lby(cU,1s,e.tt.in.,gcU,nW,cUU−−l1,1,U.−.=.l,cU−1{U,n(−x,l.⊤1.,..,.c.U,−xl⊤n,1),⊤.|.x.i,cU−l,∈n)}. dj k+1 dj j+1 dj k This gives dimR(VU−l) = 2nk − dimR(WU−l) = lutevaluesofcoefficientsontherighthandsidearesmalleror 2nk − 2n2(l + 1) = 2n(k − nl − n). And again we equalthatone.InthiscaseletK =(K \π(c ))∪{π(c )}. 2 1 j k+1 find C such that U−l−1 It is clear that K is a basis for U. 2 forNoalwl isim=ila1r,ly2,f.o.r.m,na−nkewanbdaswisriKtei+K1 u=sinKgnb−aks+is1 K=i πU−l(CU−l−1)=O(2n(k−nl−nvuu) NU2Nn−(2lk−n−k1nl−n))=O(NU−−kl−n−ln+1l−nn). {π(c ),π(c ),...,π(c )}. t U−l−1 j1 j2 jk whNeroew|dπl,(ic|l)≤=ndflo,1rπa(cllj1l)a+nddli,2πbe(ccaj2u)se+e·v·e·ry+tdiml,keπw(chjekn) BL=em(mπa(C2.1)⊤g,.iv.e.s,πth(aCt if )A⊤,C=⊤)⊤(Ct1⊤h,e.n..d,eCt(U⊤A)A⊤†)an=d we took some π(ch) off from the basis it then had a such det(BB†2) th1at is of sizUe U−1 U representation in the new basis that all the coordinates were absolutelysmallerorequalthanzero.Repeatingthisprocedure O((U−2N−k−nln+l−nn)2n)=O(U−1N−k2−n2n((UU−−ll))). at most n−k ≤ n times and using triangle inequality gives Y U−l−1 Y l thentheproperty.Henceifv=a c +a c +···+a c with l=0 l=1 1 1 2 2 n n |a |≤ N then π(v) = b π(c )+b π(c )+···+b π(c ) i 1 j1 2 j2 k jk with |b |≤n2N for all i. i Theorem 2.3: For a MIMO-MAC lattice code REFERENCES (L1(N1),L2(N2),...,LU(NU))ofU users,eachtransmitting [1] H.-F.Lu,J.Lahtonen, R. Vehkalahti, and C.Hollanti: Remarks onthe withntransmissionantennas,havingacodeoflengthk ≥Un, criteriaofconstructingMAC-DMToptimalcodespresentedattheIEEE Inf.TheoryWorkshop,Cairo,Egypt,2010. and each users lattice is of full rank r = 2kn we have a [2] J.Lahtonen,R.Vehkalahti,H.-F.Lu,C.Hollanti,andE.Viterbo:Onthe constant K such that DecayoftheDeterminantsofMultiuserMIMOLatticeCodespresented attheIEEEInf.TheoryWorkshop,Cairo, Egypt,Jan.2010. D(N ,...,N )≤KU−1N−k2−n2n((UU−−ll)) 1 U Y l l=1

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.