sdrangel/plugins/channelrx/demoddatv/leansdr/rs.h

#ifndef LEANSDR_RS_H
#define LEANSDR_RS_H

#include "leansdr/math.h"

#define DEBUG_RS 0

namespace leansdr
{

// Finite group GF(2^N).

// GF(2) is the ring ({0,1},+,*).
// GF(2)[X] is the ring of polynomials with coefficients in GF(2).
// P(X) is an irreducible polynomial of GF(2)[X].
// N is the degree of P(x).
// P is the bitfield representation of P(X), with degree 0 at LSB.
// GF(2)[X]/(P) is GF(2)[X] modulo P(X).
// (GF(2)[X]/(P), +) is a group with 2^N elements.
// (GF(2)[X]/(P)*, *) is a group with 2^N-1 elements.
// (GF(2)[X]/(P), +, *) is a field with 2^N elements, noted GF(2^N).
// Te is a C++ integer type for representing elements of GF(2^N).
//   "0" is 0
//   "1" is 1
//   "2" is X
//   "3" is X+1
//   "4" is X^2
// Tp is a C++ integer type for representing P(X) (1 bit larger than Te).
// ALPHA is a primitive element of GF(2^N). Usually "2"=[X] is chosen.

template<typename Te, typename Tp, Tp P, int N, Te ALPHA>
struct gf2x_p
{
    gf2x_p()
    {
        if (ALPHA != 2)
        {
            fail("gf2x_p::gf2x_p", "alpha!=2 not implemented");
            return;
        }
        // Precompute log and exp tables.
        Tp alpha_i = 1;
        for (Tp i = 0; i < (1 << N); ++i)
        {
            lut_exp[i] = alpha_i;
            lut_exp[((1 << N) - 1) + i] = alpha_i;
            lut_log[alpha_i] = i;
            alpha_i <<= 1;  // Multiply by alpha=[X] i.e. increase degrees
            if (alpha_i & (1 << N))
                alpha_i ^= P;  // Modulo P iteratively
        }
    }
    static const Te alpha = ALPHA;
    inline Te add(Te x, Te y)
    {
        return x ^ y;
    }  // Addition modulo 2
    inline Te sub(Te x, Te y)
    {
        return x ^ y;
    }  // Subtraction modulo 2
    inline Te mul(Te x, Te y)
    {
        if (!x || !y)
            return 0;
        return lut_exp[lut_log[x] + lut_log[y]];
    }
    inline Te div(Te x, Te y)
    {
        //if ( ! y ) fail("div"); // TODO
        if (!x)
            return 0;
        return lut_exp[lut_log[x] + ((1 << N) - 1) - lut_log[y]];
    }
    inline Te inv(Te x)
    {
        //    if ( ! x ) fail("inv");
        return lut_exp[((1 << N) - 1) - lut_log[x]];
    }
    inline Te exp(Te x)
    {
        return lut_exp[x];
    }
    inline Te log(Te x)
    {
        return lut_log[x];
    }
private:
    Te lut_exp[(1 << N) * 2];  // Wrap to avoid indexing modulo 2^N-1
    Te lut_log[1 << N];
};

// Reed-Solomon for RS(204,188) shortened from RS(255,239).

struct rs_engine
{
    // EN 300 421, section 4.4.2, Field Generator Polynomial
    // p(X) = X^8 + X^4 + X^3 + X^2 + 1
    gf2x_p<unsigned char, unsigned short, 0x11d, 8, 2> gf;

    u8 G[17];  // { G_16, ..., G_0 }

    rs_engine()
    {
        // EN 300 421, section 4.4.2, Code Generator Polynomial
        // G(X) = (X-alpha^0)*...*(X-alpha^15)
        for (int i = 0; i <= 16; ++i)
            G[i] = (i == 16) ? 1 : 0;  // Init G=1
        for (int d = 0; d < 16; ++d)
        {
            // Multiply by (X-alpha^d)
            // G := X*G - alpha^d*G
            for (int i = 0; i <= 16; ++i)
                G[i] = gf.sub((i == 16) ? 0 : G[i + 1], gf.mul(gf.exp(d), G[i]));
        }
#if DEBUG_RS
        fprintf(stderr, "RS generator:");
        for ( int i=0; i<=16; ++i ) fprintf(stderr, " %02x", G[i]);
        fprintf(stderr, "\n");
#endif
    }

    // RS-encoded messages are interpreted as coefficients in
    // GF(256) of a polynomial P.
    // The syndromes are synd[i] = P(alpha^i).
    // By convention coefficients are listed by decreasing degree here,
    // so we can evaluate syndromes of the shortened code without
    // prepending with 51 zeroes.
    bool syndromes(const u8 *poly, u8 *synd)
    {
        bool corrupted = false;
        for (int i = 0; i < 16; ++i)
        {
            synd[i] = eval_poly_rev(poly, 204, gf.exp(i));
            if (synd[i])
                corrupted = true;
        }
        return corrupted;
    }
    u8 eval_poly_rev(const u8 *poly, int n, u8 x)
    {
        // poly[0]*x^(n-1) + .. + poly[n-1]*x^0 with Hörner method.
        u8 acc = 0;
        for (int i = 0; i < n; ++i)
            acc = gf.add(gf.mul(acc, x), poly[i]);
        return acc;
    }

    // Evaluation with coefficients listed by increasing degree.
    u8 eval_poly(const u8 *poly, int deg, u8 x)
    {
        // poly[0]*x^0 + .. + poly[deg]*x^deg with Hörner method.
        u8 acc = 0;
        for (; deg >= 0; --deg)
            acc = gf.add(gf.mul(acc, x), poly[deg]);
        return acc;
    }

    // Append parity symbols

    void encode(u8 msg[204])
    {
        // TBD Avoid copying
        u8 p[204];
        memcpy(p, msg, 204); // was 188 but causing underflow (PVS https://www.viva64.com/en/w/v512/)
        memset(p + 188, 0, 16);
        // p = msg*X^16
#if DEBUG_RS
        fprintf(stderr, "uncoded:");
        for ( int i=0; i<204; ++i ) fprintf(stderr, " %d", p[i]);
        fprintf(stderr, "\n");
#endif
        // Compute remainder modulo G
        for (int d = 0; d < 188; ++d)
        {
            // Clear monomial of degree d
            if (!p[d])
                continue;
            u8 k = gf.div(p[d], G[0]);
            // p(X) := p(X) - k*G(X)*X^(188-d)
            for (int i = 0; i <= 16; ++i)
                p[d + i] = gf.sub(p[d + i], gf.mul(k, G[i]));
        }
#if DEBUG_RS
        fprintf(stderr, "coded:");
        for ( int i=0; i<204; ++i ) fprintf(stderr, " %d", p[i]);
        fprintf(stderr, "\n");
#endif
        memcpy(msg + 188, p + 188, 16);
    }

    // Try to fix errors in pout[].
    // If pin[] is provided, errors will be fixed in the original
    // message too and syndromes will be updated.

    bool correct(u8 synd[16], u8 pout[188], u8 pin[204] = NULL, int *bits_corrected = NULL)
    {
        // Berlekamp - Massey
        // http://en.wikipedia.org/wiki/Berlekamp%E2%80%93Massey_algorithm#Code_sample
        u8 C[16] =
        { 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };  // Max degree is L
        u8 B[16] =
        { 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
        int L = 0;
        int m = 1;
        u8 b = 1;
        for (int n = 0; n < 16; ++n)
        {
            u8 d = synd[n];
            for (int i = 1; i <= L; ++i)
                d ^= gf.mul(C[i], synd[n - i]);
            if (!d)
            {
                ++m;
            }
            else if (2 * L <= n)
            {
                u8 T[16];
                memcpy(T, C, sizeof(T));
                for (int i = 0; i < 16 - m; ++i)
                    C[m + i] ^= gf.mul(d, gf.mul(gf.inv(b), B[i]));
                L = n + 1 - L;
                memcpy(B, T, sizeof(B));
                b = d;
                m = 1;
            }
            else
            {
                for (int i = 0; i < 16 - m; ++i)
                    C[m + i] ^= gf.mul(d, gf.mul(gf.inv(b), B[i]));
                ++m;
            }
        }
        // L is the number of errors
        // C of degree L is now the error locator polynomial (Lambda)
#if DEBUG_RS
        fprintf(stderr, "[L=%d  C=",L);
        for ( int i=0; i<16; ++i ) fprintf(stderr, " %d", C[i]);
        fprintf(stderr, "]\n");
        fprintf(stderr, "[S=");
        for ( int i=0; i<16; ++i ) fprintf(stderr, " %d", synd[i]);
        fprintf(stderr, "]\n");
#endif

        // Forney
        // http://en.wikipedia.org/wiki/Forney_algorithm  (2t=16)

        // Compute Omega
        u8 omega[16];
        memset(omega, 0, sizeof(omega));
        // TODO loops
        for (int i = 0; i < 16; ++i)
            for (int j = 0; j < 16; ++j)
                if (i + j < 16)
                    omega[i + j] ^= gf.mul(synd[i], C[j]);
#if DEBUG_RS
        fprintf(stderr, "omega=");
        for ( int i=0; i<16; ++i ) fprintf(stderr, " %d", omega[i]);
        fprintf(stderr, "\n");
#endif

        // Compute Lambda'
        u8 Cprime[15];
        for (int i = 0; i < 15; ++i)
            Cprime[i] = (i & 1) ? 0 : C[i + 1];
#if DEBUG_RS
        fprintf(stderr, "Cprime=");
        for ( int i=0; i<15; ++i ) fprintf(stderr, " %d", Cprime[i]);
        fprintf(stderr, "\n");
#endif

        // Find zeroes of C by exhaustive search?
        // TODO Chien method
        int roots_found = 0;
        for (int i = 0; i < 255; ++i)
        {
            u8 r = gf.exp(i);  // Candidate root alpha^0..alpha^254
            u8 v = eval_poly(C, L, r);
            if (!v)
            {
                // r is a root X_k^-1 of the error locator polynomial.
                u8 xk = gf.inv(r);
                int loc = (255 - i) % 255;  // == log(xk)
#if DEBUG_RS
                        fprintf(stderr, "found root=%d, inv=%d, loc=%d\n", r, xk, loc);
#endif
                if (loc < 204)
                {
                    // Evaluate e_k
                    u8 num = gf.mul(xk, eval_poly(omega, L, r));
                    u8 den = eval_poly(Cprime, 14, r);
                    u8 e = gf.div(num, den);
                    // Subtract e from coefficient of degree loc.
                    // Note: Coeffients listed by decreasing degree in pin[] and pout[].
                    if (bits_corrected)
                        *bits_corrected += hamming_weight(e);
                    if (loc >= 16)
                        pout[203 - loc] ^= e;
                    if (pin)
                        pin[203 - loc] ^= e;
                }
                if (++roots_found == L)
                    break;
            }
        }
        if (pin)
            return syndromes(pin, synd);
        else
            return false;
    }

};

}  // namespace

#endif  // LEANSDR_RS_H