The length of a typical Huffman codeword

If p/sub i/(i=1,/spl middot//spl middot//spl middot/, N) is the probability of the ith letter of a memoryless source, the length l/sub i/ of the corresponding binary Huffman codeword can be very different from the value -log p/sub i/. For a typical letter, however, l/sub i//spl ap/-logp/sub i/. More precisely, P/sub m//sup -/=/spl Sigma//sub j/spl isin/{i|l<-logpj-m}/p/sub j/<2/sup -m/ and P/sub m//sup +/=/spl Sigma//sub j/spl isin/{i|li>-logpi+m/}p/sub j/<2/sup -c(m-2)+2/, where c/spl ap/2.27.<
>