(a) Let x∈RN and define y∈R2N by
yn={xn,x2N−n−1,0⩽n⩽N−1N⩽n⩽2N−1
Let Y∈C2N be defined as the discrete Fourier transform (DFT) of y, i.e.
Yk=n=0∑2N−1ynω2Nnk,ω2N=exp(−πi/N),0⩽k⩽2N−1
Show that
Yk=2ω2N−k/2n=0∑N−1xncos[Nπ(n+21)k],0⩽k⩽2N−1
(b) Define the discrete cosine transform (DCT)CN:RN→RN by
z=CNx, where zk=n=0∑N−1xncos[Nπ(n+21)k],k=0,…,N−1
For N=2p with p∈N, show that, similar to the Fast Fourier Transform (FFT), there exists an algorithm that computes the DCT of a vector of length N, where the number of multiplications required is bounded by CNlogN, where C is some constant independent of N.
[You may not assume that the FFT algorithm requires O(NlogN) multiplications to compute the DFT of a vector of length N. If you use this, you must prove it.]