diff --git a/src/engine/platform/nes.cpp b/src/engine/platform/nes.cpp
index 733ba2630..aae2d4ff0 100644
--- a/src/engine/platform/nes.cpp
+++ b/src/engine/platform/nes.cpp
@@ -62,7 +62,7 @@ const char** DivPlatformNES::getRegisterSheet() {
   return regCheatSheetNES;
 }
 
-void DivPlatformNES::doWrite(unsigned short addr, unsigned char data) {
+void DivPlatformNES::doWrite(int ts, unsigned short addr, unsigned char data) {
   if (useNP) {
     if (isE) {
       e1_NP->Write(addr,data);
@@ -72,13 +72,13 @@ void DivPlatformNES::doWrite(unsigned short addr, unsigned char data) {
       nes2_NP->Write(addr,data);
     }
   } else {
-    apu_wr_reg(nes,addr,data);
+    apu_wr_reg(nes,ts,addr,data);
   }
 }
 
 #define doPCM \
   if (!dpcmMode && dacSample!=-1) { \
-    dacPeriod+=dacRate; \
+    dacPeriod+=dacRate*pcmAdvance; \
     if (dacPeriod>=rate) { \
       DivSample* s=parent->getSample(dacSample); \
       if (s->samples>0 && dacPos<s->samples) { \
@@ -86,10 +86,10 @@ void DivPlatformNES::doWrite(unsigned short addr, unsigned char data) {
           unsigned char next=((unsigned char)s->data8[dacPos]+0x80)>>1; \
           if (dacAntiClickOn && dacAntiClick<next) { \
             dacAntiClick+=8; \
-            doWrite(0x4011,dacAntiClick); \
+            doWrite(i,0x4011,dacAntiClick); \
           } else { \
             dacAntiClickOn=false; \
-            doWrite(0x4011,next); \
+            doWrite(i,0x4011,next); \
           } \
         } \
         dacPos++; \
@@ -105,30 +105,43 @@ void DivPlatformNES::doWrite(unsigned short addr, unsigned char data) {
     } \
   }
 
-void DivPlatformNES::acquire_puNES(short** buf, size_t len) {
+void DivPlatformNES::acquire_puNES(blip_buffer_t** bb, size_t len) {
   for (int i=0; i<5; i++) {
     oscBuf[i]->begin(len);
   }
 
+  nes->timestamp=0;
+  nes->bb=bb[0];
+
   for (size_t i=0; i<len; i++) {
-    doPCM;
+    // heuristic
+    int pcmAdvance=1;
+    if (writes.empty()) {
+      if (dpcmMode || dacSample==-1) {
+        break;
+      } else {
+        pcmAdvance=len-i;
+        if (dacPeriod>0) {
+          int remainTime=(rate-dacPeriod+dacRate-1)/dacRate;
+          if (remainTime<pcmAdvance) pcmAdvance=remainTime;
+          if (remainTime<1) pcmAdvance=1;
+        }
+      }
+    }
 
     if (!writes.empty()) {
+      pcmAdvance=1;
       QueuedWrite w=writes.front();
-      doWrite(w.addr,w.val);
+      doWrite(i,w.addr,w.val);
       regPool[w.addr&0x1f]=w.val;
       writes.pop();
     }
+
+    i+=pcmAdvance-1;
+
+    doPCM;
   
-    apu_tick(nes,NULL);
-    nes->apu.odd_cycle=!nes->apu.odd_cycle;
-    if (nes->apu.clocked) {
-      nes->apu.clocked=false;
-    }
-    int sample=(pulse_output(nes)+tnd_output(nes))<<6;
-    if (sample>32767) sample=32767;
-    if (sample<-32768) sample=-32768;
-    buf[0][i]=sample;
+    /*
     if (++writeOscBuf>=32) {
       writeOscBuf=0;
       oscBuf[0]->putSample(i,isMuted[0]?0:(nes->S1.output<<11));
@@ -136,8 +149,9 @@ void DivPlatformNES::acquire_puNES(short** buf, size_t len) {
       oscBuf[2]->putSample(i,isMuted[2]?0:(nes->TR.output<<11));
       oscBuf[3]->putSample(i,isMuted[3]?0:(nes->NS.output<<11));
       oscBuf[4]->putSample(i,isMuted[4]?0:(nes->DMC.output<<8));
-    }
+    }*/
   }
+  apu_tick(nes,len);
 
   for (int i=0; i<5; i++) {
     oscBuf[i]->end(len);
@@ -147,6 +161,7 @@ void DivPlatformNES::acquire_puNES(short** buf, size_t len) {
 void DivPlatformNES::acquire_NSFPlay(short** buf, size_t len) {
   int out1[2];
   int out2[2];
+  const int pcmAdvance=1;
 
   for (int i=0; i<5; i++) {
     oscBuf[i]->begin(len);
@@ -157,7 +172,7 @@ void DivPlatformNES::acquire_NSFPlay(short** buf, size_t len) {
 
     if (!writes.empty()) {
       QueuedWrite w=writes.front();
-      doWrite(w.addr,w.val);
+      doWrite(i,w.addr,w.val);
       regPool[w.addr&0x1f]=w.val;
       writes.pop();
     }
@@ -190,6 +205,7 @@ void DivPlatformNES::acquire_NSFPlay(short** buf, size_t len) {
 void DivPlatformNES::acquire_NSFPlayE(short** buf, size_t len) {
   int out1[2];
   int out2[2];
+  const int pcmAdvance=1;
 
   for (int i=0; i<5; i++) {
     oscBuf[i]->begin(len);
@@ -200,7 +216,7 @@ void DivPlatformNES::acquire_NSFPlayE(short** buf, size_t len) {
 
     if (!writes.empty()) {
       QueuedWrite w=writes.front();
-      doWrite(w.addr,w.val);
+      doWrite(i,w.addr,w.val);
       regPool[w.addr&0x1f]=w.val;
       writes.pop();
     }
@@ -231,17 +247,19 @@ void DivPlatformNES::acquire_NSFPlayE(short** buf, size_t len) {
 }
 
 void DivPlatformNES::acquire(short** buf, size_t len) {
-  if (useNP) {
-    if (isE) {
-      acquire_NSFPlayE(buf,len);
-    } else {
-      acquire_NSFPlay(buf,len);
-    }
+  if (!useNP) return;
+  if (isE) {
+    acquire_NSFPlayE(buf,len);
   } else {
-    acquire_puNES(buf,len);
+    acquire_NSFPlay(buf,len);
   }
 }
 
+void DivPlatformNES::acquireDirect(blip_buffer_t** bb, size_t len) {
+  if (useNP) return;
+  acquire_puNES(bb,len);
+}
+
 static unsigned char noiseTable[253]={
   6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 4,
   15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4,
@@ -920,6 +938,10 @@ bool DivPlatformNES::keyOffAffectsArp(int ch) {
   return true;
 }
 
+bool DivPlatformNES::hasAcquireDirect() {
+  return (!useNP && !isE);
+}
+
 void DivPlatformNES::setFlags(const DivConfig& flags) {
   int clockSel=flags.getInt("clockSel",0);
   if (clockSel==2) { // Dendy
diff --git a/src/engine/platform/nes.h b/src/engine/platform/nes.h
index 99abdebea..7c0a10aeb 100644
--- a/src/engine/platform/nes.h
+++ b/src/engine/platform/nes.h
@@ -85,14 +85,15 @@ class DivPlatformNES: public DivDispatch {
   friend void putDispatchChip(void*,int);
   friend void putDispatchChan(void*,int,int);
 
-  void doWrite(unsigned short addr, unsigned char data);
+  void doWrite(int ts, unsigned short addr, unsigned char data);
   unsigned char calcDPCMRate(int inRate);
-  void acquire_puNES(short** buf, size_t len);
+  void acquire_puNES(blip_buffer_t** bb, size_t len);
   void acquire_NSFPlay(short** buf, size_t len);
   void acquire_NSFPlayE(short** buf, size_t len);
 
   public:
     void acquire(short** buf, size_t len);
+    void acquireDirect(blip_buffer_t** bb, size_t len);
     int dispatch(DivCommand c);
     void* getChanState(int chan);
     DivMacroInt* getChanMacroInt(int ch);
@@ -104,6 +105,7 @@ class DivPlatformNES: public DivDispatch {
     void tick(bool sysTick=true);
     void muteChannel(int ch, bool mute);
     bool keyOffAffectsArp(int ch);
+    bool hasAcquireDirect();
     float getPostAmp();
     unsigned char readDMC(unsigned short addr);
     void setNSFPlay(bool use);
diff --git a/src/engine/platform/sound/nes/apu.c b/src/engine/platform/sound/nes/apu.c
index 7930c57e7..aeb083bc8 100644
--- a/src/engine/platform/sound/nes/apu.c
+++ b/src/engine/platform/sound/nes/apu.c
@@ -21,268 +21,316 @@
 #include <string.h>
 #include "apu.h"
 
-void apu_tick(struct NESAPU* a, BYTE *hwtick) {
-  /* sottraggo il numero di cicli eseguiti */
-  a->apu.cycles--;
-  /*
-   * questo flag sara' a TRUE solo nel ciclo
-   * in cui viene eseguito il length counter.
-   */
-  a->apu.length_clocked = FALSE;
-  /*
-   * se e' settato il delay del $4017, essendo
-   * questo il ciclo successivo, valorizzo il
-   * registro.
-   */
+void apu_tick(struct NESAPU* a, int len) {
+  if (len<=a->timestamp) return;
+
+  int rem=len-a->timestamp;
+  if (rem>1) {
+    // output now just in case
+    int sample=(pulse_output(a)+tnd_output(a))<<6;
+    if (sample!=a->lastSample) {  
+      blip_add_delta(a->bb,a->timestamp,sample-a->lastSample);
+      a->lastSample=sample;
+    }
+  }
+
+  while (rem>0) {
+    // predict advance
+    int advance=rem;
+    if (a->r4017.jitter.delay) {
+      advance=1;
+    }
+    if (advance>a->apu.cycles) {
+      advance=a->apu.cycles;
+    }
+    if (advance>a->S1.frequency) {
+      advance=a->S1.frequency;
+    }
+    if (advance>a->S2.frequency) {
+      advance=a->S2.frequency;
+    }
+    if (advance>a->TR.frequency) {
+      advance=a->TR.frequency;
+    }
+    if (advance>a->NS.frequency) {
+      advance=a->NS.frequency;
+    }
+    if (advance>a->DMC.frequency) {
+      advance=a->DMC.frequency;
+    }
+    if (advance<1) advance=1;
+
+    /* sottraggo il numero di cicli eseguiti */
+    a->apu.cycles-=advance;
+    /*
+    * questo flag sara' a TRUE solo nel ciclo
+    * in cui viene eseguito il length counter.
+    */
+    a->apu.length_clocked = FALSE;
+    /*
+    * se e' settato il delay del $4017, essendo
+    * questo il ciclo successivo, valorizzo il
+    * registro.
+    */
 #if defined (VECCHIA_GESTIONE_JITTER)
-  if (r4017.jitter.delay) {
-    r4017.jitter.delay = FALSE;
-    r4017_jitter();
-  }
+    if (r4017.jitter.delay) {
+      r4017.jitter.delay = FALSE;
+      r4017_jitter();
+    }
 #else
-  if (a->r4017.jitter.delay) {
-    a->r4017.jitter.delay = FALSE;
-    r4017_jitter(0)
-  }
-  r4017_reset_frame()
+    if (a->r4017.jitter.delay) {
+      a->r4017.jitter.delay = FALSE;
+      r4017_jitter(0)
+    }
+    r4017_reset_frame()
 #endif
 
-  /* quando apu.cycles e' a 0 devo eseguire uno step */
-  if (!a->apu.cycles) {
-    switch (a->apu.step) {
-      case 0:
-        /*
-         * nel mode 1 devo eseguire il
-         * length counter e lo sweep.
-         */
-        if (a->apu.mode == APU_48HZ) {
-          length_clock()
-          sweep_clock()
-        }
-        envelope_clock()
-        /* triangle's linear counter */
-        linear_clock()
-        /* passo al prossimo step */
-        apu_change_step(++a->apu.step);
-        break;
-      case 1:
-        /* nel mode 0 devo eseguire il length counter */
-        if (a->apu.mode == APU_60HZ) {
-          length_clock()
-          sweep_clock()
-        }
-        envelope_clock()
-        /* triangle's linear counter */
-        linear_clock()
-        /* passo al prossimo step */
-        apu_change_step(++a->apu.step);
-        break;
-      case 2:
-        /*
-         * nel mode 1 devo eseguire il
-         * length counter e lo sweep.
-         */
-        if (a->apu.mode == APU_48HZ) {
-          length_clock()
-          sweep_clock()
-        }
-        envelope_clock()
-        /* triangle's linear counter */
-        linear_clock()
-        /* passo al prossimo step */
-        apu_change_step(++a->apu.step);
-        break;
-      case 3:
-        /*
-         * gli step 3, 4 e 5 settano il bit 6 del $4015
-         * ma solo nel 4 genero un IRQ.
-         */
-        if (a->apu.mode == APU_60HZ) {
+    /* quando apu.cycles e' a 0 devo eseguire uno step */
+    if (!a->apu.cycles) {
+      switch (a->apu.step) {
+        case 0:
           /*
-           * se e' a 0 il bit 6 del $4017 (interrupt
-           * inhibit flag) allora devo generare un IRQ.
-           */
-          if (!(a->r4017.value & 0x40)) {
-            /* setto il bit 6 del $4015 */
-            a->r4015.value |= 0x40;
+          * nel mode 1 devo eseguire il
+          * length counter e lo sweep.
+          */
+          if (a->apu.mode == APU_48HZ) {
+            length_clock()
+            sweep_clock()
           }
-        } else {
-          /* nel mode 1 devo eseguire l'envelope */
           envelope_clock()
           /* triangle's linear counter */
           linear_clock()
-        }
-        /* passo al prossimo step */
-        apu_change_step(++a->apu.step);
-        break;
-      case 4:
-        /*
-         * gli step 3, 4 e 5 settano il bit 6 del $4015
-         * ma solo nel 4 genero un IRQ.
-         */
-        if (a->apu.mode == APU_60HZ) {
-          length_clock()
-          sweep_clock()
+          /* passo al prossimo step */
+          apu_change_step(++a->apu.step);
+          break;
+        case 1:
+          /* nel mode 0 devo eseguire il length counter */
+          if (a->apu.mode == APU_60HZ) {
+            length_clock()
+            sweep_clock()
+          }
           envelope_clock()
           /* triangle's linear counter */
           linear_clock()
+          /* passo al prossimo step */
+          apu_change_step(++a->apu.step);
+          break;
+        case 2:
           /*
-           * se e' a 0 il bit 6 del $4017 (interrupt
-           * inhibit flag) allora devo generare un IRQ.
-           */
-          if (!(a->r4017.value & 0x40)) {
-            /* setto il bit 6 del $4015 */
-            a->r4015.value |= 0x40;
+          * nel mode 1 devo eseguire il
+          * length counter e lo sweep.
+          */
+          if (a->apu.mode == APU_48HZ) {
+            length_clock()
+            sweep_clock()
           }
-        }
-        /* passo al prossimo step */
-        apu_change_step(++a->apu.step);
-        break;
-      case 5:
-        /*
-         * gli step 3, 4 e 5 settano il bit 6 del $4015
-         * ma solo nel 4 genero un IRQ.
-         */
-        if (a->apu.mode == APU_60HZ) {
+          envelope_clock()
+          /* triangle's linear counter */
+          linear_clock()
+          /* passo al prossimo step */
+          apu_change_step(++a->apu.step);
+          break;
+        case 3:
           /*
-           * se e' a 0 il bit 6 del $4017 (interrupt
-           * inhibit flag) allora devo generare un IRQ.
-           */
-          if (!(a->r4017.value & 0x40)) {
-            /* setto il bit 6 del $4015 */
-            a->r4015.value |= 0x40;
+          * gli step 3, 4 e 5 settano il bit 6 del $4015
+          * ma solo nel 4 genero un IRQ.
+          */
+          if (a->apu.mode == APU_60HZ) {
+            /*
+            * se e' a 0 il bit 6 del $4017 (interrupt
+            * inhibit flag) allora devo generare un IRQ.
+            */
+            if (!(a->r4017.value & 0x40)) {
+              /* setto il bit 6 del $4015 */
+              a->r4015.value |= 0x40;
+            }
+          } else {
+            /* nel mode 1 devo eseguire l'envelope */
+            envelope_clock()
+            /* triangle's linear counter */
+            linear_clock()
+          }
+          /* passo al prossimo step */
+          apu_change_step(++a->apu.step);
+          break;
+        case 4:
+          /*
+          * gli step 3, 4 e 5 settano il bit 6 del $4015
+          * ma solo nel 4 genero un IRQ.
+          */
+          if (a->apu.mode == APU_60HZ) {
+            length_clock()
+            sweep_clock()
+            envelope_clock()
+            /* triangle's linear counter */
+            linear_clock()
+            /*
+            * se e' a 0 il bit 6 del $4017 (interrupt
+            * inhibit flag) allora devo generare un IRQ.
+            */
+            if (!(a->r4017.value & 0x40)) {
+              /* setto il bit 6 del $4015 */
+              a->r4015.value |= 0x40;
+            }
+          }
+          /* passo al prossimo step */
+          apu_change_step(++a->apu.step);
+          break;
+        case 5:
+          /*
+          * gli step 3, 4 e 5 settano il bit 6 del $4015
+          * ma solo nel 4 genero un IRQ.
+          */
+          if (a->apu.mode == APU_60HZ) {
+            /*
+            * se e' a 0 il bit 6 del $4017 (interrupt
+            * inhibit flag) allora devo generare un IRQ.
+            */
+            if (!(a->r4017.value & 0x40)) {
+              /* setto il bit 6 del $4015 */
+              a->r4015.value |= 0x40;
+            }
+            a->apu.step++;
+          } else {
+            /* nel mode 1 devo ricominciare il ciclo */
+            a->apu.step = 0;
+          }
+          /* passo al prossimo step */
+          apu_change_step(a->apu.step);
+          break;
+        case 6:
+          /* da qui ci passo solo nel mode 0 */
+          envelope_clock()
+          /* triangle's linear counter */
+          linear_clock()
+          /* questo e' il passaggio finale del mode 0 */
+          a->apu.step = 1;
+          /* passo al prossimo step */
+          apu_change_step(a->apu.step);
+          break;
+      }
+    }
+
+    /*
+    * eseguo un ticket per ogni canale
+    * valorizzandone l'output.
+    */
+    // SQUARE 1 TICK
+    if (!(a->S1.frequency-=advance)) {
+      square_output(a->S1, 0)
+      a->S1.frequency = (a->S1.timer + 1) << 1;
+      a->S1.sequencer = (a->S1.sequencer + 1) & 0x07;
+    }
+
+    // SQUARE 2 TICK
+    if (!(a->S2.frequency-=advance)) {
+      square_output(a->S2, 0)
+      a->S2.frequency = (a->S2.timer + 1) << 1;
+      a->S2.sequencer = (a->S2.sequencer + 1) & 0x07;
+    }
+
+    // TRIANGLE TICK
+    if (!(a->TR.frequency-=advance)) {
+      a->TR.frequency = a->TR.timer + 1;
+      if (a->TR.length.value && a->TR.linear.value) {
+        a->TR.sequencer = (a->TR.sequencer + 1) & 0x1F;
+        triangle_output()
+      }
+    }
+
+    // NOISE TICK
+    if (!(a->NS.frequency-=advance)) {
+      if (a->NS.mode) {
+        a->NS.shift = (a->NS.shift >> 1) | (((a->NS.shift ^ (a->NS.shift >> 6)) & 0x0001) << 14);
+      } else {
+        a->NS.shift = (a->NS.shift >> 1) | (((a->NS.shift ^ (a->NS.shift >> 1)) & 0x0001) << 14);
+      }
+      a->NS.shift &= 0x7FFF;
+      noise_output()
+      a->NS.frequency = noise_timer[a->apu.type][a->NS.timer];
+    }
+
+    // DMC TICK
+    if (!(a->DMC.frequency-=advance)) {
+      if (!a->DMC.silence) {
+        if (!(a->DMC.shift & 0x01)) {
+          if (a->DMC.counter > 1) {
+            a->DMC.counter -= 2;
           }
-          a->apu.step++;
         } else {
-          /* nel mode 1 devo ricominciare il ciclo */
-          a->apu.step = 0;
+          if (a->DMC.counter < 126) {
+            a->DMC.counter += 2;
+          }
         }
-        /* passo al prossimo step */
-        apu_change_step(a->apu.step);
-        break;
-      case 6:
-        /* da qui ci passo solo nel mode 0 */
-        envelope_clock()
-        /* triangle's linear counter */
-        linear_clock()
-        /* questo e' il passaggio finale del mode 0 */
-        a->apu.step = 1;
-        /* passo al prossimo step */
-        apu_change_step(a->apu.step);
-        break;
-    }
-  }
-
-  /*
-   * eseguo un ticket per ogni canale
-   * valorizzandone l'output.
-   */
-  // SQUARE 1 TICK
-  if (!(--a->S1.frequency)) {
-    square_output(a->S1, 0)
-    a->S1.frequency = (a->S1.timer + 1) << 1;
-    a->S1.sequencer = (a->S1.sequencer + 1) & 0x07;
-    a->apu.clocked = TRUE;
-  }
-
-  // SQUARE 2 TICK
-  if (!(--a->S2.frequency)) {
-    square_output(a->S2, 0)
-    a->S2.frequency = (a->S2.timer + 1) << 1;
-    a->S2.sequencer = (a->S2.sequencer + 1) & 0x07;
-    a->apu.clocked = TRUE;
-  }
-
-  // TRIANGLE TICK
-  if (!(--a->TR.frequency)) {
-    a->TR.frequency = a->TR.timer + 1;
-    if (a->TR.length.value && a->TR.linear.value) {
-      a->TR.sequencer = (a->TR.sequencer + 1) & 0x1F;
-      triangle_output()
-      a->apu.clocked = TRUE;
-    }
-  }
-
-  // NOISE TICK
-  if (!(--a->NS.frequency)) {
-    if (a->NS.mode) {
-      a->NS.shift = (a->NS.shift >> 1) | (((a->NS.shift ^ (a->NS.shift >> 6)) & 0x0001) << 14);
-    } else {
-      a->NS.shift = (a->NS.shift >> 1) | (((a->NS.shift ^ (a->NS.shift >> 1)) & 0x0001) << 14);
-    }
-    a->NS.shift &= 0x7FFF;
-    noise_output()
-    a->NS.frequency = noise_timer[a->apu.type][a->NS.timer];
-    a->apu.clocked = TRUE;
-  }
-
-  // DMC TICK
-  if (!(--a->DMC.frequency)) {
-    if (!a->DMC.silence) {
-      if (!(a->DMC.shift & 0x01)) {
-        if (a->DMC.counter > 1) {
-          a->DMC.counter -= 2;
+      }
+      a->DMC.shift >>= 1;
+      dmc_output();
+      if (!(--a->DMC.counter_out)) {
+        a->DMC.counter_out = 8;
+        if (!a->DMC.empty) {
+          a->DMC.shift = a->DMC.buffer;
+          a->DMC.empty = TRUE;
+          a->DMC.silence = FALSE;
+        } else {
+          a->DMC.silence = TRUE;
         }
-      } else {
-        if (a->DMC.counter < 126) {
-          a->DMC.counter += 2;
+      }
+      a->DMC.frequency = dmc_rate[a->apu.type][a->DMC.rate_index];
+    }
+    if (a->DMC.empty && a->DMC.remain) {
+      BYTE tick = 4;
+      switch (a->DMC.tick_type) {
+        case DMC_CPU_WRITE:
+          tick = 3;
+          break;
+        case DMC_R4014:
+          tick = 2;
+          break;
+        case DMC_NNL_DMA:
+          tick = 1;
+          break;
+      }
+      {
+        a->DMC.buffer = a->readDMC(a->readDMCUser,a->DMC.address);
+      }
+      /* e naturalmente incremento anche quelli eseguiti dall'opcode */
+      a->apu.cpu_cycles += tick;
+      /* salvo a che ciclo dell'istruzione avviene il dma */
+      a->DMC.dma_cycle = a->apu.cpu_opcode_cycle;
+      /* il DMC non e' vuoto */
+      a->DMC.empty = FALSE;
+      if (++a->DMC.address > 0xFFFF) {
+        a->DMC.address = 0x8000;
+      }
+      if (!(--a->DMC.remain)) {
+        if (a->DMC.loop) {
+          a->DMC.remain = a->DMC.length;
+          a->DMC.address = a->DMC.address_start;
+        } else if (a->DMC.irq_enabled) {
+          a->r4015.value |= 0x80;
         }
       }
     }
-    a->DMC.shift >>= 1;
-    dmc_output();
-    if (!(--a->DMC.counter_out)) {
-      a->DMC.counter_out = 8;
-      if (!a->DMC.empty) {
-        a->DMC.shift = a->DMC.buffer;
-        a->DMC.empty = TRUE;
-        a->DMC.silence = FALSE;
-      } else {
-        a->DMC.silence = TRUE;
-      }
-    }
-    a->DMC.frequency = dmc_rate[a->apu.type][a->DMC.rate_index];
-    a->apu.clocked = TRUE;
-  }
-  if (a->DMC.empty && a->DMC.remain) {
-    BYTE tick = 4;
-    switch (a->DMC.tick_type) {
-      case DMC_CPU_WRITE:
-        tick = 3;
-        break;
-      case DMC_R4014:
-        tick = 2;
-        break;
-      case DMC_NNL_DMA:
-        tick = 1;
-        break;
-    }
-    {
-      a->DMC.buffer = a->readDMC(a->readDMCUser,a->DMC.address);
-    }
-    /* incremento gli hwtick da compiere */
-    if (hwtick) { hwtick[0] += tick; }
-    /* e naturalmente incremento anche quelli eseguiti dall'opcode */
-    a->apu.cpu_cycles += tick;
-    /* salvo a che ciclo dell'istruzione avviene il dma */
-    a->DMC.dma_cycle = a->apu.cpu_opcode_cycle;
-    /* il DMC non e' vuoto */
-    a->DMC.empty = FALSE;
-    if (++a->DMC.address > 0xFFFF) {
-      a->DMC.address = 0x8000;
-    }
-    if (!(--a->DMC.remain)) {
-      if (a->DMC.loop) {
-        a->DMC.remain = a->DMC.length;
-        a->DMC.address = a->DMC.address_start;
-      } else if (a->DMC.irq_enabled) {
-        a->r4015.value |= 0x80;
-      }
-    }
-  }
 
-  a->r4011.cycles++;
+    a->r4011.cycles+=advance;
+    if (advance&1) {
+      a->apu.odd_cycle=!a->apu.odd_cycle;
+    }
+
+    // output sample
+    a->timestamp+=advance-1;
+    int sample=(pulse_output(a)+tnd_output(a))<<6;
+    if (sample!=a->lastSample) {  
+      blip_add_delta(a->bb,a->timestamp,sample-a->lastSample);
+      a->lastSample=sample;
+    }
+    
+    rem-=advance;
+    a->timestamp++;
+  }
+  a->timestamp=len;
 }
+
 void apu_turn_on(struct NESAPU* a, BYTE apu_type) {
   memset(&a->apu, 0x00, sizeof(a->apu));
   memset(&a->r4015, 0x00, sizeof(a->r4015));
@@ -322,4 +370,5 @@ void apu_turn_on(struct NESAPU* a, BYTE apu_type) {
   a->apu.odd_cycle = 0;
   // come non viene inizializzato? Vorrei qualche spiegazione...
   a->r4011.frames = 0;
+  a->lastSample = 0;
 }
diff --git a/src/engine/platform/sound/nes/apu.h b/src/engine/platform/sound/nes/apu.h
index 1e1afffae..b4da93ead 100644
--- a/src/engine/platform/sound/nes/apu.h
+++ b/src/engine/platform/sound/nes/apu.h
@@ -22,6 +22,7 @@
 #define APU_H_
 
 #include "common.h"
+#include "blip_buf.h"
 
 enum dmc_types_of_dma { DMC_NORMAL, DMC_CPU_WRITE, DMC_R4014, DMC_NNL_DMA };
 enum apu_channels { APU_S1, APU_S2, APU_TR, APU_NS, APU_DMC, APU_EXTRA, APU_MASTER };
@@ -277,12 +278,6 @@ typedef struct _apu {
   int cpu_cycles;
   int cpu_opcode_cycle;
   BYTE odd_cycle;
-
-  /* ------------------------------------------------------- */
-  /* questi valori non e' necessario salvarli nei savestates */
-  /* ------------------------------------------------------- */
-  /* */ BYTE clocked;                                     /* */
-  /* ------------------------------------------------------- */
 } _apu;
 typedef struct _r4011 {
   BYTE value;
@@ -433,8 +428,11 @@ EXTERNC struct NESAPU {
   _apuTriangle TR;
   _apuNoise NS;
   _apuDMC DMC;
+  blip_buffer_t* bb;
   void* readDMCUser;
   unsigned char (*readDMC)(void*,unsigned short);
+  int timestamp;
+  int lastSample;
   unsigned char muted[5];
 };
 
@@ -540,7 +538,7 @@ static const WORD dmc_rate[3][16] = {
   }
 };
 
-EXTERNC void apu_tick(struct NESAPU* a, BYTE *hwtick);
+EXTERNC void apu_tick(struct NESAPU* a, int len);
 EXTERNC void apu_turn_on(struct NESAPU* a, BYTE apu_type);
 
 #undef EXTERNC
diff --git a/src/engine/platform/sound/nes/cpu_inline.h b/src/engine/platform/sound/nes/cpu_inline.h
index 87ea3c9ee..8fd903df4 100644
--- a/src/engine/platform/sound/nes/cpu_inline.h
+++ b/src/engine/platform/sound/nes/cpu_inline.h
@@ -39,7 +39,9 @@
     r2006.value++;\
   }
 
-INLINE static void apu_wr_reg(struct NESAPU* a, WORD address, BYTE value) {
+INLINE static void apu_wr_reg(struct NESAPU* a, int ts, WORD address, BYTE value) {
+  apu_tick(a,ts);
+
   if (!(address & 0x0010)) {
     /* -------------------- square 1 --------------------*/
     if (address <= 0x4003) {
@@ -200,7 +202,6 @@ INLINE static void apu_wr_reg(struct NESAPU* a, WORD address, BYTE value) {
         } else {
           a->DMC.counter = a->DMC.output = value;
         }
-        a->apu.clocked = TRUE;
 
         a->r4011.cycles = a->r4011.frames = 0;
         a->r4011.value = value;