path: root/libs/fluidsynth/src/fluid_iir_filter.c

/* FluidSynth - A Software Synthesizer
 *
 * Copyright (C) 2003  Peter Hanappe and others.
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public License
 * as published by the Free Software Foundation; either version 2.1 of
 * the License, or (at your option) any later version.
 *
 * This library 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free
 * Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
 * 02110-1301, USA
 */

#include "fluid_iir_filter.h"
#include "fluid_sys.h"
#include "fluid_conv.h"

/**
 * Applies a low- or high-pass filter with variable cutoff frequency and quality factor
 * for a given biquad transfer function:
 *          b0 + b1*z^-1 + b2*z^-2
 *  H(z) = ------------------------
 *          a0 + a1*z^-1 + a2*z^-2
 *
 * Also modifies filter state accordingly.
 * @param iir_filter Filter parameter
 * @param dsp_buf Pointer to the synthesized audio data
 * @param count Count of samples in dsp_buf
 */
/*
 * Variable description:
 * - dsp_a1, dsp_a2: Filter coefficients for the the previously filtered output signal
 * - dsp_b0, dsp_b1, dsp_b2: Filter coefficients for input signal
 * - coefficients normalized to a0
 *
 * A couple of variables are used internally, their results are discarded:
 * - dsp_i: Index through the output buffer
 * - dsp_centernode: delay line for the IIR filter
 * - dsp_hist1: same
 * - dsp_hist2: same
 */
void
fluid_iir_filter_apply(fluid_iir_filter_t *iir_filter,
                       fluid_real_t *dsp_buf, int count)
{
    if(iir_filter->type == FLUID_IIR_DISABLED || iir_filter->q_lin == 0)
    {
        return;
    }
    else
    {
        /* IIR filter sample history */
        fluid_real_t dsp_hist1 = iir_filter->hist1;
        fluid_real_t dsp_hist2 = iir_filter->hist2;

        /* IIR filter coefficients */
        fluid_real_t dsp_a1 = iir_filter->a1;
        fluid_real_t dsp_a2 = iir_filter->a2;
        fluid_real_t dsp_b02 = iir_filter->b02;
        fluid_real_t dsp_b1 = iir_filter->b1;
        int dsp_filter_coeff_incr_count = iir_filter->filter_coeff_incr_count;

        fluid_real_t dsp_centernode;
        int dsp_i;

        /* filter (implement the voice filter according to SoundFont standard) */

        /* Check for denormal number (too close to zero). */
        if(FLUID_FABS(dsp_hist1) < 1e-20f)
        {
            dsp_hist1 = 0.0f;    /* FIXME JMG - Is this even needed? */
        }

        /* Two versions of the filter loop. One, while the filter is
        * changing towards its new setting. The other, if the filter
        * doesn't change.
        */

        if(dsp_filter_coeff_incr_count > 0)
        {
            fluid_real_t dsp_a1_incr = iir_filter->a1_incr;
            fluid_real_t dsp_a2_incr = iir_filter->a2_incr;
            fluid_real_t dsp_b02_incr = iir_filter->b02_incr;
            fluid_real_t dsp_b1_incr = iir_filter->b1_incr;


            /* Increment is added to each filter coefficient filter_coeff_incr_count times. */
            for(dsp_i = 0; dsp_i < count; dsp_i++)
            {
                /* The filter is implemented in Direct-II form. */
                dsp_centernode = dsp_buf[dsp_i] - dsp_a1 * dsp_hist1 - dsp_a2 * dsp_hist2;
                dsp_buf[dsp_i] = dsp_b02 * (dsp_centernode + dsp_hist2) + dsp_b1 * dsp_hist1;
                dsp_hist2 = dsp_hist1;
                dsp_hist1 = dsp_centernode;

                if(dsp_filter_coeff_incr_count-- > 0)
                {
                    fluid_real_t old_b02 = dsp_b02;
                    dsp_a1 += dsp_a1_incr;
                    dsp_a2 += dsp_a2_incr;
                    dsp_b02 += dsp_b02_incr;
                    dsp_b1 += dsp_b1_incr;

                    /* Compensate history to avoid the filter going havoc with large frequency changes */
                    if(iir_filter->compensate_incr && FLUID_FABS(dsp_b02) > 0.001f)
                    {
                        fluid_real_t compensate = old_b02 / dsp_b02;
                        dsp_hist1 *= compensate;
                        dsp_hist2 *= compensate;
                    }
                }
            } /* for dsp_i */
        }
        else /* The filter parameters are constant.  This is duplicated to save time. */
        {
            for(dsp_i = 0; dsp_i < count; dsp_i++)
            {
                /* The filter is implemented in Direct-II form. */
                dsp_centernode = dsp_buf[dsp_i] - dsp_a1 * dsp_hist1 - dsp_a2 * dsp_hist2;
                dsp_buf[dsp_i] = dsp_b02 * (dsp_centernode + dsp_hist2) + dsp_b1 * dsp_hist1;
                dsp_hist2 = dsp_hist1;
                dsp_hist1 = dsp_centernode;
            }
        }

        iir_filter->hist1 = dsp_hist1;
        iir_filter->hist2 = dsp_hist2;
        iir_filter->a1 = dsp_a1;
        iir_filter->a2 = dsp_a2;
        iir_filter->b02 = dsp_b02;
        iir_filter->b1 = dsp_b1;
        iir_filter->filter_coeff_incr_count = dsp_filter_coeff_incr_count;

        fluid_check_fpe("voice_filter");
    }
}


DECLARE_FLUID_RVOICE_FUNCTION(fluid_iir_filter_init)
{
    fluid_iir_filter_t *iir_filter = obj;
    enum fluid_iir_filter_type type = param[0].i;
    enum fluid_iir_filter_flags flags = param[1].i;

    iir_filter->type = type;
    iir_filter->flags = flags;

    if(type != FLUID_IIR_DISABLED)
    {
        fluid_iir_filter_reset(iir_filter);
    }
}

void
fluid_iir_filter_reset(fluid_iir_filter_t *iir_filter)
{
    iir_filter->hist1 = 0;
    iir_filter->hist2 = 0;
    iir_filter->last_fres = -1.;
    iir_filter->q_lin = 0;
    iir_filter->filter_startup = 1;
}

DECLARE_FLUID_RVOICE_FUNCTION(fluid_iir_filter_set_fres)
{
    fluid_iir_filter_t *iir_filter = obj;
    fluid_real_t fres = param[0].real;

    iir_filter->fres = fres;
    iir_filter->last_fres = -1.;
}

static fluid_real_t fluid_iir_filter_q_from_dB(fluid_real_t q_dB)
{
    /* The generator contains 'centibels' (1/10 dB) => divide by 10 to
     * obtain dB */
    q_dB /= 10.0f;

    /* Range: SF2.01 section 8.1.3 # 8 (convert from cB to dB => /10) */
    fluid_clip(q_dB, 0.0f, 96.0f);

    /* Short version: Modify the Q definition in a way, that a Q of 0
     * dB leads to no resonance hump in the freq. response.
     *
     * Long version: From SF2.01, page 39, item 9 (initialFilterQ):
     * "The gain at the cutoff frequency may be less than zero when
     * zero is specified".  Assume q_dB=0 / q_lin=1: If we would leave
     * q as it is, then this results in a 3 dB hump slightly below
     * fc. At fc, the gain is exactly the DC gain (0 dB).  What is
     * (probably) meant here is that the filter does not show a
     * resonance hump for q_dB=0. In this case, the corresponding
     * q_lin is 1/sqrt(2)=0.707.  The filter should have 3 dB of
     * attenuation at fc now.  In this case Q_dB is the height of the
     * resonance peak not over the DC gain, but over the frequency
     * response of a non-resonant filter.  This idea is implemented as
     * follows: */
    q_dB -= 3.01f;

    /* The 'sound font' Q is defined in dB. The filter needs a linear
       q. Convert. */
    return FLUID_POW(10.0f, q_dB / 20.0f);
}

DECLARE_FLUID_RVOICE_FUNCTION(fluid_iir_filter_set_q)
{
    fluid_iir_filter_t *iir_filter = obj;
    fluid_real_t q = param[0].real;
    int flags = iir_filter->flags;

    if(flags & FLUID_IIR_Q_ZERO_OFF && q <= 0.0)
    {
        q = 0;
    }
    else if(flags & FLUID_IIR_Q_LINEAR)
    {
        /* q is linear (only for user-defined filter)
         * increase to avoid Q being somewhere between zero and one,
         * which results in some strange amplified lowpass signal
         */
        q++;
    }
    else
    {
        q = fluid_iir_filter_q_from_dB(q);
    }

    iir_filter->q_lin = q;
    iir_filter->filter_gain = 1.0;

    if(!(flags & FLUID_IIR_NO_GAIN_AMP))
    {
        /* SF 2.01 page 59:
         *
         *  The SoundFont specs ask for a gain reduction equal to half the
         *  height of the resonance peak (Q).  For example, for a 10 dB
         *  resonance peak, the gain is reduced by 5 dB.  This is done by
         *  multiplying the total gain with sqrt(1/Q).  `Sqrt' divides dB
         *  by 2 (100 lin = 40 dB, 10 lin = 20 dB, 3.16 lin = 10 dB etc)
         *  The gain is later factored into the 'b' coefficients
         *  (numerator of the filter equation).  This gain factor depends
         *  only on Q, so this is the right place to calculate it.
         */
        iir_filter->filter_gain /= FLUID_SQRT(q);
    }

    /* The synthesis loop will have to recalculate the filter coefficients. */
    iir_filter->last_fres = -1.;
}

static FLUID_INLINE void
fluid_iir_filter_calculate_coefficients(fluid_iir_filter_t *iir_filter,
                                        int transition_samples,
                                        fluid_real_t output_rate)
{
    /* FLUID_IIR_Q_LINEAR may switch the filter off by setting Q==0 */
    if(iir_filter->q_lin == 0)
    {
        return;
    }
    else
    {
        /*
         * Those equations from Robert Bristow-Johnson's `Cookbook
         * formulae for audio EQ biquad filter coefficients', obtained
         * from Harmony-central.com / Computer / Programming. They are
         * the result of the bilinear transform on an analogue filter
         * prototype. To quote, `BLT frequency warping has been taken
         * into account for both significant frequency relocation and for
         * bandwidth readjustment'. */

        fluid_real_t omega = (fluid_real_t)(2.0 * M_PI) *
                                            (iir_filter->last_fres / output_rate);
        fluid_real_t sin_coeff = FLUID_SIN(omega);
        fluid_real_t cos_coeff = FLUID_COS(omega);
        fluid_real_t alpha_coeff = sin_coeff / (2.0f * iir_filter->q_lin);
        fluid_real_t a0_inv = 1.0f / (1.0f + alpha_coeff);

        /* Calculate the filter coefficients. All coefficients are
         * normalized by a0. Think of `a1' as `a1/a0'.
         *
         * Here a couple of multiplications are saved by reusing common expressions.
         * The original equations should be:
         *  iir_filter->b0=(1.-cos_coeff)*a0_inv*0.5*iir_filter->filter_gain;
         *  iir_filter->b1=(1.-cos_coeff)*a0_inv*iir_filter->filter_gain;
         *  iir_filter->b2=(1.-cos_coeff)*a0_inv*0.5*iir_filter->filter_gain; */

        /* "a" coeffs are same for all 3 available filter types */
        fluid_real_t a1_temp = -2.0f * cos_coeff * a0_inv;
        fluid_real_t a2_temp = (1.0f - alpha_coeff) * a0_inv;

        fluid_real_t b02_temp, b1_temp;

        switch(iir_filter->type)
        {
        case FLUID_IIR_HIGHPASS:
            b1_temp = (1.0f + cos_coeff) * a0_inv * iir_filter->filter_gain;

            /* both b0 -and- b2 */
            b02_temp = b1_temp * 0.5f;

            b1_temp *= -1.0f;
            break;

        case FLUID_IIR_LOWPASS:
            b1_temp = (1.0f - cos_coeff) * a0_inv * iir_filter->filter_gain;

            /* both b0 -and- b2 */
            b02_temp = b1_temp * 0.5f;
            break;

        default:
            /* filter disabled, should never get here */
            return;
        }

        iir_filter->compensate_incr = 0;

        if(iir_filter->filter_startup || (transition_samples == 0))
        {
            /* The filter is calculated, because the voice was started up.
             * In this case set the filter coefficients without delay.
             */
            iir_filter->a1 = a1_temp;
            iir_filter->a2 = a2_temp;
            iir_filter->b02 = b02_temp;
            iir_filter->b1 = b1_temp;
            iir_filter->filter_coeff_incr_count = 0;
            iir_filter->filter_startup = 0;
//       printf("Setting initial filter coefficients.\n");
        }
        else
        {

            /* The filter frequency is changed.  Calculate an increment
             * factor, so that the new setting is reached after one buffer
             * length. x_incr is added to the current value FLUID_BUFSIZE
             * times. The length is arbitrarily chosen. Longer than one
             * buffer will sacrifice some performance, though.  Note: If
             * the filter is still too 'grainy', then increase this number
             * at will.
             */

            iir_filter->a1_incr = (a1_temp - iir_filter->a1) / transition_samples;
            iir_filter->a2_incr = (a2_temp - iir_filter->a2) / transition_samples;
            iir_filter->b02_incr = (b02_temp - iir_filter->b02) / transition_samples;
            iir_filter->b1_incr = (b1_temp - iir_filter->b1) / transition_samples;

            if(FLUID_FABS(iir_filter->b02) > 0.0001f)
            {
                fluid_real_t quota = b02_temp / iir_filter->b02;
                iir_filter->compensate_incr = quota < 0.5f || quota > 2.f;
            }

            /* Have to add the increments filter_coeff_incr_count times. */
            iir_filter->filter_coeff_incr_count = transition_samples;
        }

        fluid_check_fpe("voice_write filter calculation");
    }
}


void fluid_iir_filter_calc(fluid_iir_filter_t *iir_filter,
                           fluid_real_t output_rate,
                           fluid_real_t fres_mod)
{
    fluid_real_t fres;

    /* calculate the frequency of the resonant filter in Hz */
    fres = fluid_ct2hz(iir_filter->fres + fres_mod);

    /* FIXME - Still potential for a click during turn on, can we interpolate
       between 20khz cutoff and 0 Q? */

    /* I removed the optimization of turning the filter off when the
     * resonance frequence is above the maximum frequency. Instead, the
     * filter frequency is set to a maximum of 0.45 times the sampling
     * rate. For a 44100 kHz sampling rate, this amounts to 19845
     * Hz. The reason is that there were problems with anti-aliasing when the
     * synthesizer was run at lower sampling rates. Thanks to Stephan
     * Tassart for pointing me to this bug. By turning the filter on and
     * clipping the maximum filter frequency at 0.45*srate, the filter
     * is used as an anti-aliasing filter. */

    if(fres > 0.45f * output_rate)
    {
        fres = 0.45f * output_rate;
    }
    else if(fres < 5.f)
    {
        fres = 5.f;
    }

    /* if filter enabled and there is a significant frequency change.. */
    if(iir_filter->type != FLUID_IIR_DISABLED && FLUID_FABS(fres - iir_filter->last_fres) > 0.01f)
    {
        /* The filter coefficients have to be recalculated (filter
         * parameters have changed). Recalculation for various reasons is
         * forced by setting last_fres to -1.  The flag filter_startup
         * indicates, that the DSP loop runs for the first time, in this
         * case, the filter is set directly, instead of smoothly fading
         * between old and new settings. */
        iir_filter->last_fres = fres;
        fluid_iir_filter_calculate_coefficients(iir_filter, FLUID_BUFSIZE,
                                                output_rate);
    }


    fluid_check_fpe("voice_write DSP coefficients");

}