diff --git a/sdk/software/examples/fft_dma/fft.h b/sdk/software/examples/fft_dma/fft.h
index cf86335..03a38f7 100644
--- a/sdk/software/examples/fft_dma/fft.h
+++ b/sdk/software/examples/fft_dma/fft.h
@@ -10,6 +10,7 @@
 #define FFT_STAT_DONE     (1 << 1)
 #define FFT_STAT_BUSY     (1 << 0)
 #define FFT_POINT_NUM     1024
+#define FFT_SAMPLE_BITS    16
 
 void fft_start();
 void fft_wait();
diff --git a/sdk/software/examples/fft_dma/main.c b/sdk/software/examples/fft_dma/main.c
index 954ec0e..dec934f 100644
--- a/sdk/software/examples/fft_dma/main.c
+++ b/sdk/software/examples/fft_dma/main.c
@@ -67,7 +67,8 @@ void sw_fft(float re[], float im[], int n) {
     }
 }
 
-// 全局对齐数组：作为 DMA 的源和目的内存
+// 全局对齐数组：作为 DMA 的源和目的内存。
+// FFT 外设仍按 32-bit AXI word 编址，每个 word 的低 16 bit 作为一个 signed 样本。
 int32_t hw_in_re_arr[FFT_POINT_NUM]  __attribute__((aligned(64)));
 int32_t hw_in_im_arr[FFT_POINT_NUM]  __attribute__((aligned(64)));
 int32_t hw_out_re_arr[FFT_POINT_NUM] __attribute__((aligned(64)));
@@ -95,13 +96,13 @@ int main(int argc, char** argv)
     float sw_in_im[FFT_POINT_NUM];
 
     for (int i = 0; i < FFT_POINT_NUM; i++) {
-        int32_t sample = fft_test_sample((unsigned int)i);
+        int16_t sample = (int16_t)fft_test_sample((unsigned int)i);
 
         sw_in_re[i] = (float)sample;
         sw_in_im[i] = 0.0f;
 
-        // 硬件输入需要转为整数存入无缓存内存
-        uncached_in_re[i]  = sample;
+        // 硬件输入为 16-bit signed，写入 32-bit word 的低半字。
+        uncached_in_re[i]  = (int32_t)sample;
         uncached_in_im[i]  = 0;
         
         // 清理输出内存以防干扰
@@ -116,7 +117,7 @@ int main(int argc, char** argv)
     printf("\n--- Starting Hardware FFT with DMA ---\n");
     tick_start = get_ns(); // 开始计时
 
-    uint32_t transfer_bytes = FFT_POINT_NUM * 4; // 1024个点 * 4字节
+    uint32_t transfer_bytes = FFT_POINT_NUM * 4; // 1024个点 * 4字节，低16位有效
 
     // MA 将数据从内存搬运到 FFT 输入外设
     dma_start_transfer(0, ((uint32_t)hw_in_re_arr & 0x1FFFFFFF), phys_fft_in_re, transfer_bytes, 100);
@@ -158,19 +159,31 @@ int main(int argc, char** argv)
         printf("Speedup Ratio    : %.2fx\n", (float)sw_time / hw_time);
     }
 
-    printf("\n--- Verification (Only showing Bins with energy > 10) ---\n");
+    printf("\n--- Verification (%d-bit FFT, only showing bins with energy > 10) ---\n", FFT_SAMPLE_BITS);
+    int max_abs_err_re = 0;
+    int max_abs_err_im = 0;
     for (int i = 0; i < FFT_POINT_NUM; i++) {
         // CPU 通过无缓存指针读取 DMA 搬运回来的结果
         int32_t hw_re = uncached_out_re[i];
         int32_t hw_im = uncached_out_im[i];
+        int32_t sw_re = (int32_t)sw_in_re[i] / FFT_POINT_NUM;
+        int32_t sw_im = (int32_t)sw_in_im[i] / FFT_POINT_NUM;
+        int err_re = (int)(hw_re - sw_re);
+        int err_im = (int)(hw_im - sw_im);
+
+        if (err_re < 0) err_re = -err_re;
+        if (err_im < 0) err_im = -err_im;
+        if (err_re > max_abs_err_re) max_abs_err_re = err_re;
+        if (err_im > max_abs_err_im) max_abs_err_im = err_im;
 
         if (fabs((float)hw_re) > 10 || fabs((float)hw_im) > 10) {
             printf("Bin [%4d] Hz: HW(Re:%6d, Im:%6d) | SW(Re:%6d, Im:%6d)\n", 
                     i, 
                     hw_re, hw_im, 
-                    (int)sw_in_re[i] / FFT_POINT_NUM, (int)sw_in_im[i] / FFT_POINT_NUM);
+                    sw_re, sw_im);
         }
     }
+    printf("Max Abs Error  : Re=%d, Im=%d\n", max_abs_err_re, max_abs_err_im);
 
     return 0;
 }