dsp.h

Go to the documentation of this file.

// This file is part of LeanSDR Copyright (C) 2016-2018 <pabr@pabr.org>.
// See the toplevel README for more information.
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program.  If not, see <http://www.gnu.org/licenses/>.

#ifndef LEANSDR_DSP_H
#define LEANSDR_DSP_H

#include "leansdr/framework.h"
#include "leansdr/math.h"
#include <math.h>

namespace leansdr
{

// DSP blocks

// [cconverter] converts complex streams between numric types,
// with optional ofsetting and rational scaling.
template <typename Tin, int Zin, typename Tout, int Zout, int Gn, int Gd>
struct cconverter : runnable
{
    cconverter(scheduler *sch, pipebuf<complex<Tin>> &_in,
               pipebuf<complex<Tout>> &_out)
        : runnable(sch, "cconverter"),
          in(_in), out(_out)
    {
    }
    void run()
    {
        unsigned long count = min(in.readable(), out.writable());
        complex<Tin> *pin = in.rd(), *pend = pin + count;
        complex<Tout> *pout = out.wr();
        for (; pin < pend; ++pin, ++pout)
        {
            pout->re = Zout + (pin->re - (Tin)Zin) * Gn / Gd;
            pout->im = Zout + (pin->im - (Tin)Zin) * Gn / Gd;
        }
        in.read(count);
        out.written(count);
    }

  private:
    pipereader<complex<Tin>> in;
    pipewriter<complex<Tout>> out;
};

template <typename T>
struct cfft_engine
{
    const int n;
    cfft_engine(int _n) : n(_n), invsqrtn(1.0 / sqrt(n))
    {
        // Compute log2(n)
        logn = 0;
        for (int t = n; t > 1; t >>= 1)
            ++logn;
        // Bit reversal
        bitrev = new int[n];
        for (int i = 0; i < n; ++i)
        {
            bitrev[i] = 0;
            for (int b = 0; b < logn; ++b)
                bitrev[i] = (bitrev[i] << 1) | ((i >> b) & 1);
        }
        // Float constants
        omega = new complex<T>[n];
        omega_rev = new complex<T>[n];
        for (int i = 0; i < n; ++i)
        {
            float a = 2.0 * M_PI * i / n;
            omega_rev[i].re = (omega[i].re = cosf(a));
            omega_rev[i].im = -(omega[i].im = sinf(a));
        }
    }
    void inplace(complex<T> *data, bool reverse = false)
    {
        // Bit-reversal permutation
        for (int i = 0; i < n; ++i)
        {
            int r = bitrev[i];
            if (r < i)
            {
                complex<T> tmp = data[i];
                data[i] = data[r];
                data[r] = tmp;
            }
        }
        complex<T> *om = reverse ? omega_rev : omega;
        // Danielson-Lanczos
        for (int i = 0; i < logn; ++i)
        {
            int hbs = 1 << i;
            int dom = 1 << (logn - 1 - i);
            for (int j = 0; j < dom; ++j)
            {
                int p = j * hbs * 2, q = p + hbs;
                for (int k = 0; k < hbs; ++k)
                {
                    complex<T> &w = om[k * dom];
                    complex<T> &dqk = data[q + k];
                    complex<T> x(w.re * dqk.re - w.im * dqk.im,
                                 w.re * dqk.im + w.im * dqk.re);
                    data[q + k].re = data[p + k].re - x.re;
                    data[q + k].im = data[p + k].im - x.im;
                    data[p + k].re = data[p + k].re + x.re;
                    data[p + k].im = data[p + k].im + x.im;
                }
            }
        }
        if (reverse)
        {
            float invn = 1.0 / n;
            for (int i = 0; i < n; ++i)
            {
                data[i].re *= invn;
                data[i].im *= invn;
            }
        }
    }

  private:
    int logn;
    int *bitrev;
    complex<T> *omega, *omega_rev;
    float invsqrtn;
};

template <typename T>
struct adder : runnable
{
    adder(scheduler *sch,
          pipebuf<T> &_in1, pipebuf<T> &_in2, pipebuf<T> &_out)
        : runnable(sch, "adder"),
          in1(_in1), in2(_in2), out(_out)
    {
    }
    void run()
    {
        int n = out.writable();
        if (in1.readable() < n)
            n = in1.readable();
        if (in2.readable() < n)
            n = in2.readable();
        T *pin1 = in1.rd(), *pin2 = in2.rd(), *pout = out.wr(), *pend = pout + n;
        while (pout < pend)
            *pout++ = *pin1++ + *pin2++;
        in1.read(n);
        in2.read(n);
        out.written(n);
    }

  private:
    pipereader<T> in1, in2;
    pipewriter<T> out;
};

template <typename Tscale, typename Tin, typename Tout>
struct scaler : runnable
{
    Tscale scale;
    scaler(scheduler *sch, Tscale _scale,
           pipebuf<Tin> &_in, pipebuf<Tout> &_out)
        : runnable(sch, "scaler"),
          scale(_scale),
          in(_in), out(_out)
    {
    }
    void run()
    {
        unsigned long count = min(in.readable(), out.writable());
        Tin *pin = in.rd(), *pend = pin + count;
        Tout *pout = out.wr();
        for (; pin < pend; ++pin, ++pout)
            *pout = *pin * scale;
        in.read(count);
        out.written(count);
    }

  private:
    pipereader<Tin> in;
    pipewriter<Tout> out;
};

// [awgb_c] generates complex white gaussian noise.

template <typename T>
struct wgn_c : runnable
{
    wgn_c(scheduler *sch, pipebuf<complex<T>> &_out)
        : runnable(sch, "awgn"), stddev(1.0), out(_out)
    {
    }
    void run()
    {
        int n = out.writable();
        complex<T> *pout = out.wr(), *pend = pout + n;
        while (pout < pend)
        {
            // TAOCP
            float x, y, r2;
            do
            {
                x = 2 * drand48() - 1;
                y = 2 * drand48() - 1;
                r2 = x * x + y * y;
            } while (r2 == 0 || r2 >= 1);
            float k = sqrtf(-logf(r2) / r2) * stddev;
            pout->re = k * x;
            pout->im = k * y;
            ++pout;
        }
        out.written(n);
    }
    float stddev;

  private:
    pipewriter<complex<T>> out;
};

template <typename T>
struct naive_lowpass : runnable
{
    naive_lowpass(scheduler *sch, pipebuf<T> &_in, pipebuf<T> &_out, int _w)
        : runnable(sch, "lowpass"), in(_in), out(_out), w(_w)
    {
    }

    void run()
    {
        if (in.readable() < w)
            return;
        unsigned long count = min(in.readable() - w, out.writable());
        T *pin = in.rd(), *pend = pin + count;
        T *pout = out.wr();
        float k = 1.0 / w;
        for (; pin < pend; ++pin, ++pout)
        {
            T x = 0.0;
            for (int i = 0; i < w; ++i)
                x = x + pin[i];
            *pout = x * k;
        }
        in.read(count);
        out.written(count);
    }

  private:
    pipereader<T> in;
    pipewriter<T> out;
    int w;
};

template <typename T, typename Tc>
struct fir_filter : runnable
{
    fir_filter(scheduler *sch, int _ncoeffs, Tc *_coeffs,
               pipebuf<T> &_in, pipebuf<T> &_out,
               unsigned int _decim = 1)
        : runnable(sch, "fir_filter"),
          ncoeffs(_ncoeffs), coeffs(_coeffs),
          in(_in), out(_out),
          decim(_decim),
          freq_tap(NULL), tap_multiplier(1), freq_tol(0.1)
    {
        shifted_coeffs = new T[ncoeffs];
        set_freq(0);
    }

    void run()
    {
        if (in.readable() < ncoeffs)
            return;

        if (freq_tap)
        {
            float new_freq = *freq_tap * tap_multiplier;
            if (fabs(current_freq - new_freq) > freq_tol)
            {
                if (sch->verbose)
                    fprintf(stderr, "Shifting filter %f -> %f\n",
                            current_freq, new_freq);
                set_freq(new_freq);
            }
        }

        long count = min((in.readable() - ncoeffs) / decim,
                                  out.writable());
        T *pin = in.rd() + ncoeffs, *pend = pin + count * decim, *pout = out.wr();
        // TBD use coeffs when current_freq=0 (fewer mults if float)
        for (; pin < pend; pin += decim, ++pout)
        {
            T *pc = shifted_coeffs;
            T *pi = pin;
            T x = 0;
            for (int i = ncoeffs; i--; ++pc, --pi)
                x = x + (*pc) * (*pi);
            *pout = x;
        }
        in.read(count * decim);
        out.written(count);
    }

  private:
    int ncoeffs;
    Tc *coeffs;
    pipereader<T> in;
    pipewriter<T> out;
    int decim;

    T *shifted_coeffs;
    float current_freq;
    void set_freq(float f)
    {
        for (int i = 0; i < ncoeffs; ++i)
        {
            float a = 2 * M_PI * f * (i - ncoeffs / 2.0);
            float c = cosf(a), s = sinf(a);
            // TBD Support T=complex
            shifted_coeffs[i].re = coeffs[i] * c;
            shifted_coeffs[i].im = coeffs[i] * s;
        }
        current_freq = f;
    }

  public:
    float *freq_tap;
    float tap_multiplier;
    float freq_tol;
}; // fir_filter

// FIR FILTER WITH INTERPOLATION AND DECIMATION

template <typename T, typename Tc>
struct fir_resampler : runnable
{
    fir_resampler(scheduler *sch, int _ncoeffs, Tc *_coeffs,
                  pipebuf<T> &_in, pipebuf<T> &_out,
                  int _interp = 1, int _decim = 1)
        : runnable(sch, "fir_resampler"),
          ncoeffs(_ncoeffs), coeffs(_coeffs),
          interp(_interp), decim(_decim),
          in(_in), out(_out, interp),
          freq_tap(NULL), tap_multiplier(1), freq_tol(0.1)
    {
        if (decim != 1)
            fail("fir_resampler: decim not implemented"); // TBD
        shifted_coeffs = new T[ncoeffs];
        set_freq(0);
    }

    void run()
    {
        if (in.readable() < ncoeffs)
            return;

        if (freq_tap)
        {
            float new_freq = *freq_tap * tap_multiplier;
            if (fabs(current_freq - new_freq) > freq_tol)
            {
                if (sch->verbose)
                    fprintf(stderr, "Shifting filter %f -> %f\n",
                            current_freq, new_freq);
                set_freq(new_freq);
            }
        }

        if (in.readable() * interp < ncoeffs)
            return;
        unsigned long count = min((in.readable() * interp - ncoeffs) / interp,
                                  out.writable() / interp);
        int latency = (ncoeffs + interp) / interp;
        T *pin = in.rd() + latency, *pend = pin + count, *pout = out.wr();
        // TBD use coeffs when current_freq=0 (fewer mults if float)
        for (; pin < pend; ++pin)
        {
            for (int i = 0; i < interp; ++i, ++pout)
            {
                T *pi = pin;
                T *pc = shifted_coeffs + i, *pcend = shifted_coeffs + ncoeffs;
                T x = 0;
                for (; pc < pcend; pc += interp, --pi)
                    x = x + (*pc) * (*pi);
                *pout = x;
            }
        }
        in.read(count);
        out.written(count * interp);
    }

  private:
    unsigned int ncoeffs;
    Tc *coeffs;
    int interp, decim;
    pipereader<T> in;
    pipewriter<T> out;

  public:
    float *freq_tap;
    float tap_multiplier;
    float freq_tol;

  private:
    T *shifted_coeffs;
    float current_freq;
    void set_freq(float f)
    {
        for (int i = 0; i < ncoeffs; ++i)
        {
            float a = 2 * M_PI * f * i;
            float c = cosf(a), s = sinf(a);
            // TBD Support T=complex
            shifted_coeffs[i].re = coeffs[i] * c;
            shifted_coeffs[i].im = coeffs[i] * s;
        }
        current_freq = f;
    }
}; // fir_resampler

} // namespace leansdr

#endif // LEANSDR_DSP_H