doku newon / Mbed 2 deprecated Peach_AudioChannelDividerAndCompensator

Audio channel divider and compensator.

Dependencies:   DokuFFTPACK DokuUSBDevice DokuUSBHost DokuUSBHostWithIso mbed

Fork of Peach_AudioChannelDividerAndCompensator by doku newon

日本語の説明は後半にあります。

Outline

This is audio channel dividing filter for multi-way multi-amplifier systems with compensation for system and room response. This program uses 32768 taps of real time 6ch floating point FIR (by FFT and superimpose) for accurate compensation of low frequency. Either USB digital audio stream (USB1) or analog audio signal (LINE-IN of USB-DAC) can be accepted. GR-Peach divide up them into six channel (Low,Mid,High each L,R). Then outputs them to the multi-channel USB-DAC as a USB host (USB0). 32767 taps of FIR coefficient is given by the files in the USB MSD (USB0).
GR-Peach's superabundant process power solves real time 6ch/32kTaps floating point audio FIR in textbook method without any optimization.
/media/uploads/dokunewon/--2.jpg

Sorry

This cord works but doesn't tested completely, can work with multi-channnel USB-DAC CM6206 chip inside.

Connection

To use this program, connect USB0 of GR-Peach to a self powered USBHUB, a multi-channel USB-DAC(CM6206 chip inside / e.c.DN-USBSA6CHOPThttp://www.donya.jp/site/item/item_20087.html or RA-AUD51) and a USB mass storage device (coefficient files are saved) to the USBHUB as shown above.
If you would like to accept digital audio stream (16bit/48ksps) as a USB audio device, connect USB1 of GR-Peach to the PC. If not, analog stereo audio signal is accepted.

Filter setting

You can get your favorite cross-over and compensation characteristics modifying FIR coefficient files named Fir_Ch0.snd to Fir_Ch5.snd. Coefficient files are signed 16 bit integer raw formatted sound file (without any header) having time domain responses of the FIRs. Coefficient file have 65536 samples of filter response. but second-half portion of the file (32768 samples) must be zero to work superimpose correctly. If you want minimum phase FIR, filter impulse will be placed at the first sample of the file. If you want liner phase FIR, filter impulse will be placed at 16384 sample of the file.

Sample coefficient files

Sample coefficient files are shown below. store below files in Root folder of USB mass storage device before running the program. These are files for two way systems. Since flatness of frequency response is archived by the compensation, full range speaker works very well without tweeter. In below files, ch0 and ch1 for full range speakers, ch2 and ch3 for low range speakers (woofers), ch4 and ch5 are not used. Of cause, system compensation coefficient is differ according to speaker unit, speaker enclosure, room, and amplifier. Below samples are for my speakers and room.

Compensated coefficient file with channel divider

ch0:full range(high):left:/media/uploads/dokunewon/fir_ch0.snd
ch1:full range(high):right:/media/uploads/dokunewon/fir_ch1.snd
ch2:woofer(mid):left:/media/uploads/dokunewon/fir_ch2.snd
ch3:woofer(mid):right:/media/uploads/dokunewon/fir_ch3.snd
ch4:not_used(low):left:/media/uploads/dokunewon/fir_ch4.snd
ch4:not_used(low):right:/media/uploads/dokunewon/fir_ch5.snd

channel divider without compensation (pushing the user button on the GR-Peach, below coefficients are used)

ch0:full range(high):left:/media/uploads/dokunewon/fir_ch0nc.snd
ch1:full range(high):right:/media/uploads/dokunewon/fir_ch1nc.snd
ch2:woofer(mid):left:/media/uploads/dokunewon/fir_ch2nc.snd
ch3:woofer(mid):right:/media/uploads/dokunewon/fir_ch3nc.snd
ch4:not_used(low):left:/media/uploads/dokunewon/fir_ch4nc.snd
ch4:not_used(low):right:/media/uploads/dokunewon/fir_ch5nc.snd

Filter bypass test coefficient

/media/uploads/dokunewon/fir_bypass.snd

Gain of each Filter

/media/uploads/dokunewon/gains.txt

Measuring system and generating filter coefficient

These coefficient files are generated by a home made audio analyzer tool works on PC below.
/media/uploads/dokunewon/inversex90.zip
Latest beta release /media/uploads/dokunewon/inversex72.zip (Smallest number shows the latest one)
/media/uploads/dokunewon/inversex.jpg
Use above tool reading below manual.
/media/uploads/dokunewon/inversex84.pdf
Latest beta release /media/uploads/dokunewon/inversex72.pdf (Smallest number shows the latest one)
Using this tool, you can configure cross-over frequency and filter dimension for two or three way multi-way multi-amplifier system, measure response of your room and system, generate inverse response transfer function using unique and refined algorithm to create compensation FIR coefficient.

Improvement in frequency characteristic by compensation

Average spectrum of white noise before compensation (By 2 way system High:F77G98-6 x 1 Low:F77G98-6 x 8)
/media/uploads/dokunewon/uncompensated2.jpg
Average spectrum of white noise after compensation (By 2 way system High:F77G98-6 x 1 Low:F77G98-6 x 8)
/media/uploads/dokunewon/compensated2.jpg

Compensation example

Sorry. Some car noise from the road fronting my house recorded together.
Recorded playback sound by 2 way system (High:F77G98-6 x 1 Low:F77G98-6 x 8)
Early part : normal playback / latter part : compensated playback.
/media/uploads/dokunewon/samplef77g98-6.mp3
Recorded playback sound by low-cost powered small speaker Roland MA-8 (channel divider is not used)
Early part : normal playback / latter part : compensated playback.
/media/uploads/dokunewon/playbacksample.mp3
/media/uploads/dokunewon/ma8.jpg

概要

システムの特性補償を行なうマルチアンプマルチウェイシステム用のオーディオチャンネルデバイダです。低音を正確に補正するために、３２７６８タップのＦＩＲを使用しています。ＧＲ－Ｐｅａｃｈの有り余る処理力で、教科書通りのフィルタアルゴリズムが（最適化なしで！）リアルタイムにオーディオ信号を処理します。
本ソフトはＵＳＢ１端子に入力されたデジタルステレオ音声信号（４８ｋサンプル１６ビットの）か、ＵＳＢ－ＤＡＣのＬＩＮＥ－ＩＮ端子に入力されたアナログステレオ音声信号を、システム特性を補償すると同時に左右各３チャンネルに帯域分割して、マルチチャンネルＵＳＢ－ＤＡＣに出力します。

おわび

このプログラムは一応動くのですが完全にテストされていません。ＵＳＢ－ＤＡＣは内部にＣＭ６２０６を使用したものだけに対応しています。
/media/uploads/dokunewon/--2.jpg

接続

上の図のように、ＵＳＢ０には（できればセルフパワーの）ＵＳＢＨＵＢを接続し、ＨＵＢには、マルチチャンネルＵＳＢＤＡＣ（ＣＭ６２０６を使用した、ＤＮ－ＵＳＢＳＡ６ＣＨＯＰＴhttp://www.donya.jp/site/item/item_20087.html あるいは RA-AUD51）とＦＩＲ特性ファイルを記録したＵＳＢメモリを接続します。
デジタルオーディオ信号を入力する場合は、ＵＳＢ１端子にＰＣ等を接続します。ＵＳＢ１端子に何も繋がなければ、ＵＳＢ－ＤＡＣのＬＩＮＥ－ＩＮ端子のアナログオーディオ信号が入力されます。

フィルタの設定

ＦＩＲ特性ファイル（Fir_Ch0.snd ～ Fir_Ch5.snd）を変更することで、好みの補正やクロスオーバー特性を得ることができます。特性ファイルはＦＩＲの時間領域の応答を記録した１６ビット符号付整数形式のヘッダなしモノラルＲＡＷ波形データです。特性ファイルには６５５３６サンプルのフィルタ応答が記録されていますが、後ろ半分は重畳処理の都合でゼロとします。

見本の特性ファイル

見本の特性ファイルを次に示します。これらは、２ウェイシステム用のファイルです。これらのファイルはＵＳＢＭＳＤのルートフォルダに配置します。システム特性の補正によってフラットな周波数特性が実現されるので、トゥイータ無しのフルレンジスピーカーでも良い結果が得られます。チャンネル０と１がフルレンジ、２と３がウーファ、４と５は今回使用しません。もちろんシステムの補償はユニット、エンクロジャー、部屋、アンプ等によって変化します。下の例は、私のスピーカーと部屋のためのものです。（フルレンジは、F77G98-6左右各１本、ウーファはF77G98-6左右各８本を使用しています）

チャンネルデバイダの特性（補償付き）

ch0:full range(high):left:/media/uploads/dokunewon/fir_ch0.snd
ch1:full range(high):right:/media/uploads/dokunewon/fir_ch1.snd
ch2:woofer(mid):left:/media/uploads/dokunewon/fir_ch2.snd
ch3:woofer(mid):right:/media/uploads/dokunewon/fir_ch3.snd
ch4:not_used(low):left:/media/uploads/dokunewon/fir_ch4.snd
ch4:not_used(low):right:/media/uploads/dokunewon/fir_ch5.snd

チャンネルデバイダの特性（補償なし：ピーチの押しボタンを押すと、こちらの特性に切り替わります）

ch0:full range(high):left:/media/uploads/dokunewon/fir_ch0nc.snd
ch1:full range(high):right:/media/uploads/dokunewon/fir_ch1nc.snd
ch2:woofer(mid):left:/media/uploads/dokunewon/fir_ch2nc.snd
ch3:woofer(mid):right:/media/uploads/dokunewon/fir_ch3nc.snd
ch4:not_used(low):left:/media/uploads/dokunewon/fir_ch4nc.snd
ch4:not_used(low):right:/media/uploads/dokunewon/fir_ch5nc.snd

テスト用のバイパスフィルタ

/media/uploads/dokunewon/fir_bypass.snd

各フィルタのゲイン

/media/uploads/dokunewon/gains.txt

システムの測定とフィルタ特性の生成

上記の特性ファイルは次の自作のＰＣ用測定分析ツールで生成しています。
/media/uploads/dokunewon/inversex90.zip
最新のベータ版 /media/uploads/dokunewon/inversex72.zip （数字が小さい方が最新です） /media/uploads/dokunewon/inversex.jpg
使い方は次のファイルを参照してください。
/media/uploads/dokunewon/inversex84.pdf
最新のベータ版 /media/uploads/dokunewon/inversex72.pdf （数字が小さい方が最新です）
このツールを使用して、クロスオーバーの周波数や次数、システム特性の測定、補償用の逆フィルタの生成が可能です。

補償による周波数特性の改善

補償前のホワイトノイズの平均スペクトラム (By 2 way system High:F77G98-6 x 1 Low:F77G98-6 x 8)
/media/uploads/dokunewon/uncompensated2.jpg
補償後のホワイトノイズの平均スペクトラム (By 2 way system High:F77G98-6 x 1 Low:F77G98-6 x 8)
/media/uploads/dokunewon/compensated2.jpg

補償による音声の変化

国道を通る自動車の音なども一緒に入ってますが、悪しからず。
２ウェイマルチアンプシステム (High:F77G98-6 x 1 Low:mid:F77G98-6 x 8)
前半：通常再生音／後半：補償再生音
/media/uploads/dokunewon/samplef77g98-6.mp3
低価格小型パワードスピーカー Roland MA-8
前半：通常再生音／後半：補償再生音
/media/uploads/dokunewon/playbacksample.mp3
/media/uploads/dokunewon/ma8.jpg

Diff: main.cpp

Revision:
11:064c590a51f9
Parent:
10:5e85f83e8e8f
Child:
12:d12709edf53d
diff -r 5e85f83e8e8f -r 064c590a51f9 main.cpp
--- a/main.cpp	Sun Oct 18 14:23:16 2015 +0000
+++ b/main.cpp	Mon Oct 19 00:27:51 2015 +0000
@@ -26,11 +26,11 @@
 #define USB_HOST_CH     0
 #endif
 
-#define IN_DWORDS               32768                   //data size in long word buffer 32767 sample stereo
+#define IN_DWORDS               32768                       //data size in long word buffer 32767 sample stereo
 #define IN_BANKS                2
-#define OUT_DWORDS              (3 * IN_DWORDS)          //data size in long word buffer 32767 sample stereo * 3 
+#define OUT_DWORDS              (3 * IN_DWORDS)             //data size in long word buffer 32767 sample stereo * 3 
 #define OUT_BANKS               2
-#define NULL_DWORDS             (5 * 8 * 48 * 2 * 6 / 4)    //5 ITD full of data to make buffer margin
+#define NULL_DWORDS             (48 * 2 * 6 * 100 / 4)      //silenc sound data for 100msec
 
 extern int USB_Audio_Attached;
 static volatile uint32_t BufR[IN_BANKS][IN_DWORDS];
@@ -55,41 +55,66 @@
 
 Timer Tmr;
 
+#define TARGET_BUFFER_MARGIN    20          //88 for buffer full at 11 ITDs
+#define OUT_BYTE_PER_MILISEC    (48*2*6)    //48ksps 2byte 6ch / milisec
+
 void UsbSend(void const* arg) 
 {
-    int cnt;
+    int Bank = 0;
+    int Latency,Adjmsec;
     USBHostDac * p_usbdac = (USBHostDac *)arg;
+    int SkipBytes;
 
     Tmr.start();
     for(;;) 
     {
-        //wait FFT result prepaired
-        semW0.wait();
-        //if interval is too long, data shortage have happen
-        if((cnt = Tmr.read_ms()) > 695) 
+        //reset timer to measure waiting time of semahore
+        Tmr.reset();
+        if(! Bank)
+        {
+            //wait FFT result prepaired
+            semW0.wait();
+        }
+        else
+        {
+            //wait FFT result prepaired
+            semW1.wait();
+        }
+        Latency = Tmr.read_ms();
+        printf("Latency:%d\n",Latency);
+        //if semahore's waiting time is too long , data shortage have happen
+        if(Latency > 70) 
         {
+            Adjmsec = Latency - 40;
+            //it is not nessesary to fill over ITD buffer size
+            if(Adjmsec > 40) Adjmsec = 40;
+            printf("Recover %dmsec\n",Adjmsec);
             //put null data into sending queue to prevent data shortage
-            printf("Recovered by count %d\n",cnt);
-            p_usbdac->send((uint8_t *)BufNull, NULL_DWORDS * 4, false);
+            p_usbdac->send((uint8_t *)BufNull, Adjmsec * OUT_BYTE_PER_MILISEC, false);
+            //no demand for data skip because buffer have some vacancy
+            SkipBytes = 0;
         }
-        Tmr.reset();
+        //if semahore's waiting time is too short, buffer is nearly full
+        else if(Latency < 10)
+        {
+            Adjmsec = 40 - Latency;
+            if(Adjmsec > 40) Adjmsec = 40;
+            //decrease sending data, to make some vacancy on the buffer
+            SkipBytes = Adjmsec * OUT_BYTE_PER_MILISEC;
+            printf("Skip %dmsec %dByte\n",Adjmsec,SkipBytes);
+        }
+        else
+        {
+            //no demand for data skip because buffer have regular vacancy
+            SkipBytes = 0;
+        }
         LedB = 1;
         //send music data
-        p_usbdac->send((uint8_t *)BufW[0], OUT_DWORDS * 4, false);
+        p_usbdac->send((uint8_t *)&BufW[Bank][SkipBytes / 4], OUT_DWORDS * 4 - SkipBytes, false);
         LedB = 0;
-
-        //wait FFT result prepaired
-        semW1.wait();
-        //if interval is too long, data shortage have happen
-        if((cnt = Tmr.read_ms()) > 695) 
-        {
-            //put null data into sending queue to prevent data shortage
-            printf("Recovered by count %d\n",cnt);
-            p_usbdac->send((uint8_t *)BufNull, NULL_DWORDS * 4, false);
-        }
-        Tmr.reset();
-        //send music data
-        p_usbdac->send((uint8_t *)BufW[1], OUT_DWORDS * 4, false);
+        //next bank
+        Bank++;
+        Bank &= 1;
     }
 }
 
@@ -146,6 +171,9 @@
     
     printf("\n\nStarted\n");
 
+    //initializing silence buffer
+    memset((uint8_t *)BufNull,0,NULL_DWORDS * 4);
+
     //USB sound device initialize
     static USBAudio audio(48000, 2, 8000, 1, 0x7180, 0x7500);