diff --git a/src/coding/ima_decoder.c b/src/coding/ima_decoder.c
new file mode 100644
index 00000000..394f3888
--- /dev/null
+++ b/src/coding/ima_decoder.c
@@ -0,0 +1,65 @@
+#include "../util.h"
+#include "ima_decoder.h"
+
+const int32_t ADPCMTable[89] = 
+
+{
+
+    7, 8, 9, 10, 11, 12, 13, 14,
+    16, 17, 19, 21, 23, 25, 28, 31,
+    34, 37, 41, 45, 50, 55, 60, 66,
+    73, 80, 88, 97, 107, 118, 130, 143,
+    157, 173, 190, 209, 230, 253, 279, 307,
+    337, 371, 408, 449, 494, 544, 598, 658,
+    724, 796, 876, 963, 1060, 1166, 1282, 1411,
+    1552, 1707, 1878, 2066, 2272, 2499, 2749, 3024,
+    3327, 3660, 4026, 4428, 4871, 5358, 5894, 6484,
+    7132, 7845, 8630, 9493, 10442, 11487, 12635, 13899,
+    15289, 16818, 18500, 20350, 22385, 24623, 27086, 29794,
+    32767
+
+};
+
+const int IMA_IndexTable[16] = 
+
+{
+    -1, -1, -1, -1, 2, 4, 6, 8,
+    -1, -1, -1, -1, 2, 4, 6, 8 
+};
+
+void decode_nds_ima(VGMSTREAMCHANNEL * stream, sample * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do) {
+    int i=first_sample;
+    int32_t sample_count;
+    int32_t hist1 = stream->adpcm_history1_16;
+    int step_index = stream->adpcm_step_index;
+
+    if (first_sample==0) {
+        step_index = read_16bitLE(stream->offset+2,stream->streamfile);
+        hist1 = read_16bitLE(stream->offset,stream->streamfile);
+    }
+
+    for (i=first_sample,sample_count=0; i<first_sample+samples_to_do; i++,sample_count+=channelspacing) {
+        int sample_nibble = 
+                (read_8bit(stream->offset+4+i/2,stream->streamfile) >> (i&1?4:0))&0xf;
+        int delta;
+        int step = ADPCMTable[step_index];
+
+        delta = step >> 3;
+        if (sample_nibble & 1) delta += step >> 2;
+        if (sample_nibble & 2) delta += step >> 1;
+        if (sample_nibble & 4) delta += step;
+        if (sample_nibble & 8)
+            outbuf[sample_count] = clamp16(hist1 - delta);
+        else
+            outbuf[sample_count] = clamp16(hist1 + delta);
+
+        step_index += IMA_IndexTable[sample_nibble];
+        if (step_index < 0) step_index=0;
+        if (step_index > 88) step_index=88;
+
+        hist1 = outbuf[sample_count];
+    }
+
+    stream->adpcm_history1_16 = hist1;
+    stream->adpcm_step_index = step_index;
+}
diff --git a/src/coding/ima_decoder.h b/src/coding/ima_decoder.h
new file mode 100644
index 00000000..8b691cae
--- /dev/null
+++ b/src/coding/ima_decoder.h
@@ -0,0 +1,8 @@
+#include "../vgmstream.h"
+
+#ifndef _IMA_DECODER_H
+#define _IMA_DECODER_H
+
+void decode_nds_ima(VGMSTREAMCHANNEL * stream, sample * outbuf, int channelspacing, int32_t first_sample, int32_t samples_to_do);
+
+#endif
diff --git a/src/layout/interleave.c b/src/layout/interleave.c
index 8cea2b23..1f9b5bf7 100644
--- a/src/layout/interleave.c
+++ b/src/layout/interleave.c
@@ -5,25 +5,34 @@ void render_vgmstream_interleave(sample * buffer, int32_t sample_count, VGMSTREA
     int samples_written=0;
 
     int frame_size = get_vgmstream_frame_size(vgmstream);
-    int samples_per_frame = get_vgmstream_samples_per_frame(vgmstream);;
+    int samples_per_frame = get_vgmstream_samples_per_frame(vgmstream);
     int samples_this_block;
 
     samples_this_block = vgmstream->interleave_block_size / frame_size * samples_per_frame;
 
     if (vgmstream->layout_type == layout_interleave_shortblock &&
         vgmstream->current_sample - vgmstream->samples_into_block + samples_this_block> vgmstream->num_samples) {
-            samples_this_block = vgmstream->interleave_smallblock_size / frame_size * samples_per_frame;
+        frame_size = get_vgmstream_shortframe_size(vgmstream);
+        samples_per_frame = get_vgmstream_samples_per_shortframe(vgmstream);
+
+        samples_this_block = vgmstream->interleave_smallblock_size / frame_size * samples_per_frame;
     }
 
     while (samples_written<sample_count) {
         int samples_to_do; 
 
         if (vgmstream->loop_flag && vgmstream_do_loop(vgmstream)) {
-            samples_this_block = vgmstream->interleave_block_size / frame_size * samples_per_frame;
+            /* we assume that the loop is not back into a short block */
+            if (vgmstream->layout_type == layout_interleave_shortblock) {
+                frame_size = get_vgmstream_frame_size(vgmstream);
+                samples_per_frame = get_vgmstream_samples_per_frame(vgmstream);
+                samples_this_block = vgmstream->interleave_block_size / frame_size * samples_per_frame;
+            }
             continue;
         }
 
         samples_to_do = vgmstream_samples_to_do(samples_this_block, samples_per_frame, vgmstream);
+        /*printf("vgmstream_samples_to_do(samples_this_block=%d,samples_per_frame=%d,vgmstream) returns %d\n",samples_this_block,samples_per_frame,samples_to_do);*/
 
         if (samples_written+samples_to_do > sample_count)
             samples_to_do=sample_count-samples_written;
@@ -38,6 +47,8 @@ void render_vgmstream_interleave(sample * buffer, int32_t sample_count, VGMSTREA
             int chan;
             if (vgmstream->layout_type == layout_interleave_shortblock &&
                 vgmstream->current_sample + samples_this_block > vgmstream->num_samples) {
+                frame_size = get_vgmstream_shortframe_size(vgmstream);
+                samples_per_frame = get_vgmstream_samples_per_shortframe(vgmstream);
 
                 samples_this_block = vgmstream->interleave_smallblock_size / frame_size * samples_per_frame;
                 for (chan=0;chan<vgmstream->channels;chan++)
diff --git a/src/layout/nolayout.h b/src/layout/nolayout.h
index 19e400e3..4c6f1482 100644
--- a/src/layout/nolayout.h
+++ b/src/layout/nolayout.h
@@ -1,5 +1,4 @@
 /*
- * interleave.h - interleaved layouts
  */
 #include "../streamtypes.h"
 #include "../vgmstream.h"
diff --git a/src/meta/nds_strm.c b/src/meta/nds_strm.c
index 6a397c66..dc3764e5 100644
--- a/src/meta/nds_strm.c
+++ b/src/meta/nds_strm.c
@@ -70,7 +70,7 @@ VGMSTREAM * init_vgmstream_nds_strm(const char * const filename) {
     vgmstream->interleave_block_size = read_32bitLE(0x30,infile);
     vgmstream->interleave_smallblock_size = read_32bitLE(0x38,infile);
 
-    if (channel_count==1 || coding_type==coding_PCM8 || coding_type==coding_PCM16LE)
+    if (coding_type==coding_PCM8 || coding_type==coding_PCM16LE)
         vgmstream->layout_type = layout_none;
     else
         vgmstream->layout_type = layout_interleave_shortblock;
diff --git a/src/vgmstream.c b/src/vgmstream.c
index 2f2c8dfb..a38a2fa8 100644
--- a/src/vgmstream.c
+++ b/src/vgmstream.c
@@ -127,11 +127,22 @@ int get_vgmstream_samples_per_frame(VGMSTREAM * vgmstream) {
         case coding_PCM16BE:
         case coding_PCM8:
             return 1;
+        case coding_NDS_IMA:
+            return (vgmstream->interleave_block_size-4)*2;
         default:
             return 0;
     }
 }
 
+int get_vgmstream_samples_per_shortframe(VGMSTREAM * vgmstream) {
+    switch (vgmstream->coding_type) {
+        case coding_NDS_IMA:
+            return (vgmstream->interleave_smallblock_size-4)*2;
+        default:
+            return get_vgmstream_samples_per_frame(vgmstream);
+    }
+}
+
 int get_vgmstream_frame_size(VGMSTREAM * vgmstream) {
     switch (vgmstream->coding_type) {
         case coding_CRI_ADX:
@@ -143,11 +154,22 @@ int get_vgmstream_frame_size(VGMSTREAM * vgmstream) {
             return 2;
         case coding_PCM8:
             return 1;
+        case coding_NDS_IMA:
+            return vgmstream->interleave_block_size;
         default:
             return 0;
     }
 }
 
+int get_vgmstream_shortframe_size(VGMSTREAM * vgmstream) {
+    switch (vgmstream->coding_type) {
+        case coding_NDS_IMA:
+            return vgmstream->interleave_smallblock_size;
+        default:
+            return get_vgmstream_frame_size(vgmstream);
+    }
+}
+
 void decode_vgmstream(VGMSTREAM * vgmstream, int samples_written, int samples_to_do, sample * buffer) {
     int chan;
 
@@ -188,6 +210,13 @@ void decode_vgmstream(VGMSTREAM * vgmstream, int samples_written, int samples_to
                         samples_to_do);
             }
             break;
+        case coding_NDS_IMA:
+            for (chan=0;chan<vgmstream->channels;chan++) {
+                decode_nds_ima(&vgmstream->ch[chan],buffer+samples_written*vgmstream->channels+chan,
+                        vgmstream->channels,vgmstream->samples_into_block,
+                        samples_to_do);
+            }
+            break;
     }
 }
 
@@ -297,6 +326,9 @@ void describe_vgmstream(VGMSTREAM * vgmstream) {
         case coding_CRI_ADX:
             printf("CRI ADX 4-bit ADPCM");
             break;
+        case coding_NDS_IMA:
+            printf("NDS-style 4-bit IMA ADPCM");
+            break;
         default:
             printf("CANNOT DECODE");
     }
diff --git a/src/vgmstream.h b/src/vgmstream.h
index 7b71edc7..653ce996 100644
--- a/src/vgmstream.h
+++ b/src/vgmstream.h
@@ -80,6 +80,8 @@ typedef struct {
         int16_t adpcm_history2_16;  /* previous previous sample */
         int32_t adpcm_history2_32;
     };
+
+    int adpcm_step_index;     /* for IMA */
 } VGMSTREAMCHANNEL;
 
 typedef struct {
@@ -149,6 +151,9 @@ void render_vgmstream(sample * buffer, int32_t sample_count, VGMSTREAM * vgmstre
 int get_vgmstream_samples_per_frame(VGMSTREAM * vgmstream);
 /* number of bytes per frame */
 int get_vgmstream_frame_size(VGMSTREAM * vgmstream);
+/* in NDS IMA the frame size is the block size, so the last one is short */
+int get_vgmstream_samples_per_shortframe(VGMSTREAM * vgmstream);
+int get_vgmstream_shortframe_size(VGMSTREAM * vgmstream);
 
 /* Assume that we have written samples_written into the buffer already, and we have samples_to_do consecutive
  * samples ahead of us. Decode those samples into the buffer. */
diff --git a/test/Makefile b/test/Makefile
index ebb979e7..b87fb740 100644
--- a/test/Makefile
+++ b/test/Makefile
@@ -1,6 +1,6 @@
 CFLAGS=-lm -O3
 
-test: test.c ../src/streamfile.c ../src/vgmstream.c ../src/util.c ../src/meta/adx_header.c ../src/coding/adx_decoder.c ../src/coding/gcdsp_decoder.c ../src/meta/brstm.c ../src/layout/interleave.c ../src/layout/nolayout.c ../src/coding/pcm_decoder.c ../src/meta/nds_strm.c
+test: test.c ../src/streamfile.c ../src/vgmstream.c ../src/util.c ../src/meta/adx_header.c ../src/coding/adx_decoder.c ../src/coding/gcdsp_decoder.c ../src/meta/brstm.c ../src/layout/interleave.c ../src/layout/nolayout.c ../src/coding/pcm_decoder.c ../src/meta/nds_strm.c ../src/coding/ima_decoder.c
 
 clean:
 	rm test