The assignment was graded by both Cosmin and Darius.

Question 1 (Cosmin)
----------
* 18 marks (6 for 1A, 6 for 1B, 6 for 1C) are awarded as follows:
  1A (i), 1B (i), 1C (i)    3 marks for a good approximation of c(T)
  1A (ii), 1B (ii), 1C (ii)    1 marks for a good approximation of R
  1A (iii), 1B (iii), 1C (iii)    2 marks for the correct results

* 2 marks (1 for 1A, 1 for 1B) are awarded for a proper argumentation
  (including induction, or an induction-like explanation) supporting
  the relation between T and c(T). Specifically, it's not enough to
  observe (in 1A) that the first codewords have the lengths 1, 2, 3,
  4 ..., you must also explain why the n-th codeword has the length n
  (which is actually quite simple, but must not be overlooked).
  Ditto for 1B.

Total: 20 marks


Question 2 (Darius)
----------
    Q2:
         A.   2
         B.   3  (2 for the matrix and one for the entropy)
         C.   3  (2 for the matrix and one for the entropy)
         D.   3  (2 for the matrix and one for the entropy)
         E.   3  (2 for the matrix and one for the entropy)
         F.   3
         G.   3  (2 for p0 = p1 = .5, and  1 for .3651)


Total: 20 marks


Question 3 (Darius)
----------
   Q3:
         A.   6  (2 for each full code)
         B.   4  (one for each rate, and one for the comparison to H[7,4])
         C.   6  (2 for each requested disttibution)
         D.   2  (1 for saying that one bit errors are corrected;
                  1 for saying that the [4,1] code can also detect 2 bit errors)

          Here, I also counted as correct the answers were the number of errors detected was given as d-1, where d is the minimum distance (as long as the answers were consistent).


         E.   2

Total: 20 marks


Question 4 (Cosmin)
----------

* Notes:
  * If you CONSISTENTLY used the squared error for distortion (which
    gave you different final results), you will still receive proper
    credit.
  * If you started the calculations well, but you made mistakes
    on the way and obtained different (but credible) results, you
    will still receive proper credit.

* 3 marks awarded for 4A.

* 6 marks awarded for 4B.
  * 2 marks for each case (alpha = 0,30,60).
  * [The Lloyd-Max algorithm is NOT applicable]

* 6 marks awarded for 4C.
  * 2 marks for each case (alpha = 0,30,60).
  * [The Lloyd-Max algorithm IS applicable]
  * If you didn't iterate using Lloyd-Max, but you used an acceptable
    quantization method (such as equal-weight quantization), and you
    did observe that Lloyd-Max iterations can further improve the
    distortion, you will not be penalized.
  * If you didn't even mention Lloyd-Max, you will be penalized by
    1 point.

* 5 marks awarded for 4D.
  * 1 mark for alpha = 60.
  * 3 marks for alpha = 0,30.
  * 1 mark for comparing the distortion values to the results
    obtained in 4B.

Total: 20 marks

Updated: April 16, 2003