hopscotch

hopscotch

Experimental cipher using multiple hash algs for keystream.

The cipher uses multiple trusted hash algorithms, each updated on a schedule (the security factor, 1 to 10) based initially on the secret key, then on random data from a PRNG (currently MTWIST-64, also seeded from the secret key). The XOR value used to encrypt plaintext is picked from bytes of the hash outputs (being appended together into a single pool P), the hash output byte chosen used then as a modulus value to 'hop' to the next XOR value within P (hence the name 'hopscotch').

The security of the algorithm is premised on the following axioms:

1. all hash algorithms used are sufficiently unpredictable in their output based on given input (most importantly the initial key as input to the PRNG and its seeded output);
2. the PRNG is sufficiently random so as to guarantee subsequent input to all hash algorithms used as potential keystream pool bytes for XOR operations is not predictable;
3. the keystream (being the hash outputs at a given time for all hash algorithms used) is re-keyed often enough to prevent excessive re-use of bytes ('excessive' being defined by the strength parameter, restricted from 1 to 10 inclusive by empirical analysis)

Current implementation uses 3 hash algorithms, SHA512, BLAKE2B and GROESTL, each giving outputs of fixed-length = 64 bytes. Empirically, using security factors ranging from 1 to 10 (count of input bytes encrypted before re-keying by feeding 32 bytes of PRNG data to derive new hash output XOR pool P) it is unlikely that 64 picks from the pool would re-use the same bytes often enough to compromise security. Tests with the 'circle' analysis tool and the 'Tux.ppm' image test indicate ciphertext does not resemble plaintext in any obvious manner. (TODO: diehard tests or others?)

The use purely of a PRNG plus two or more already-proven hash algorithms as keystream pool material offers a simply-verified security-primitive basis for confidence, plus easy extendability by adding more hash algorithms to the keystream pool P, without complexity of a full re-analysis. So long as each individual hash algorithm is considered safe, hopping between and within the output bytes of each to derive keystream XOR material should also be safe so long as hash output bytes are not re-used extensively.

Throughput

On a modest test AMD (Linux amd_x64) encryption rates of approx. 140Mbits/s are achieved (-m 4). As this is a pure Go implementation and little effort has been put into optimization it is reasonable to expect higher rates could be achieved in the future.

$ time ./cmd -k "SuperSecret#@11ElevenTy" <blank700MB.bin >blank700MBenc.bin

real	0m40.096s
user	0m38.318s
sys	0m2.133s