Page 1 of 21 back to www.audiodesignguide.com DAC Final test June 16 st, 2006 INTRODUCTION Several years ago I started a 'long-term' project to explore the world of Digital to Analogue Conversion. Right now, after many years and a lot of testing with different technologies (see the projects DAC 1.0 to DAC 4.0 on this website and my paper published on audioxpress magazine may 2001) I imagine to have a broad vision of the sonic performance related to this critical component of the audio chain. Now it's time to explore the new components available on the market with 24bit 96-196KHz and the ASRC (Stereo Asynchronous Sample Rate Converter) that together promise an extreme sonic performances. The test will be on the best DAC chips that are the top products of 4 different manufactures: Analog Device, Wolfson, Burr-Brown and Cirrus/Crystal. The main problem using these new DAC chips is the small outline packages (28 pin SSOP or 28pin SOIC). I have found a solution to this problem using the SOIC and SSOP to DIP Adapters produced and sell by Spark Fun. On their website there is a valid guide on the SMD reflow soldering (click here)
Page 2 of 21 Follows a simple comparing table for these DAC chips. Feature Crystal Wolfson Burr- Brown DAC chip CS4397 WM8740 mono But obviously these data have no true sense. PCM1798 Burr- Brown PCM1798 mono Burr- Brown PCM1794 Burr- Brown PCM1794 mono Analog Devices AD1955 Analog Devices AD1955 mono Resolution 24 bit 24 bit 24 bit 24bit 24 bit 24bit 24 bit 24 bit Max Sampling frequency [KHz] THD @ 0dB & 44.1KHz [%] THD @ 0dB & 96KHz [%] Dynamic range 44.1KHz [db] Dynamic range 96KHz [db] 192 195 200 200 200 200 192 192-100dB -104db 0.0005 0.0005 0.0004 0.0004-123dB -110dB -100dB -104db 0.001 0.001 0.0008 0.0008-133dB -110dB 117 117 123 126 129 132 120 123 117 117 123 126 129 132 120 123 DAC S/N 100 120 123 126 129 132 120 db 120 Outputs peak to peak 1.4Vpp 2Vrms 4.0mA 8.0mA 7.8mA 15.6mA 8.64mA 17.28mA The main differences are: Only the PCM179x, AD1955 and the WF8740 can be used in mono configuration using one chip per channel. The CS4397 and WF8740 have an integrated I/V converter. The WF8740 have an internal low pass filter and the output level is 2Vrms (the optimal value for CD player) With the AD1955 probably is not possible use a passive I/V because in the datasheet Analog Devices write "Passive I/V conversion should not be used, since the DAC performance will be seriously degraded." and also because is necessary a pull-up resistor. The PCM179x and the WF8740 are very simple to set in mono because all the configuration parameters can be set in hardware mode connecting some the pin to +V or GND. The AD1955 configuration is possible only with a Microcontroller on a serial interface (SPI) and I have used for this purpose a PIC. The AD1955 have an on-chip clickless analog volume control without lost in resolution. I have used in all the measurements the Clio system by Audiomatica with the Transit by M-Audio to use the USB port of my portable pc (instead of the PCI audio board) and with the Audiophile USB by M-Audio to have a spdif ouput.
Page 3 of 21 POWER SUPPLY To build the test I have created an evaluation board for each chip and a common environment with all the necessary power supply. The DAC power supply use the LM317 and LM337 to create the +5V, -5V, +12V and - 12V. I have used a good 50W R-Core transformer get from DIY Club at a very good price. For a fine tune of output voltage I have used a Spectrol multiturn cermet trimmer (R2) with a value of 1Kohm for the 5V and 5Kohm for the 12V. Reverse diodes and capacitors polarity to get the schematic of the negative voltage. For the analog section of DAC has been used a
Page 4 of 21 shunt power supply using TL431 to get the max sonic performances (not visible in the photo). For the PCM179x and AD1955 configurated in mono mode should be used a separated shunt for each DAC chip to get the max separation from left to right channel (see schematic to the left). After the listening test the shunt regulator has been eliminated OUTPUT STAGE The DAC board has been connected to different output stages: Lundahltransformer LL7903 and LL1554 (datasheet available only of the new version). For the LL1554 and the LL7903 has been tested many connections: primaries and secondaries in series or in parallel to get the best result. The LL1554 has been excluded after the first test because the distortion at low frequency has been considered too high. CS4397 & LL1554 1000Hz and load 10Kohm pri&sec in series CS4397 & LL1554 100Hz and load 10Kohm pri&sec in series CS4397 & LL1554 1000Hz and load 10Kohm pri&sec in parallel CS4397 & LL1554 100Hz and load 10Kohm pri&sec in parallel
Page 5 of 21 The measurement with the LL7903 are perfect at any frequency. Connection n.1 from 1+1+1+1:2+2+2+2 to 1+1:2 Connection n.2 from 1+1+1+1:2+2+2+2 to 1+1:4 Connection n.3 from 1+1+1+1:2+2+2+2 to 1+1:8 INPUT STAGE (opzional) The DAC boards could be used without an input/receiver boards if you got a CD transport with I2S bus like the CD-Pro2M. Diy-high-end website Daisy For this project I have develop an input stage with all the inputs: spdif, optical toslink and USB. Using the PCM2704 no driver are necessary to use the USB Audio device with Windows XP and the most Linux distributions. The new input/receiver board use the new CS8416 with integrated input switch.
Page 6 of 21 To view the bottom of this simple board click here. OLD BOARD photo of old input board schematic of old input board The input selector does not use relay but only a high speed logic NAND chip 74HC00. The spdif signals are converted to TTL level with a circuit seen on the following website and using a high speed logic inverter 74HC04. All the input have a galvanic isolation using the Lundahl LL1572 transformer before the CS8414.
Page 7 of 21 Translation d'un signal S/PDIF vers TTL ( 0-5V ) epanorama.net - Audio Documents AD1895 / AD1896 ASRC After many time lost with Cirrus CS8421 my choice for the ASRC has been the AD1895 or the AD1896. Asynchronous sample rate conversion is converting data from one clock source at some sample rate to another clock source at the same or different sample rate. This components are very important in the Jitter reduction. There are some articles on this function on: How by Charles Altmann Peak Pro 5 Soundstage - Assemblage D2D-1 Sample-Rate Converter The input has been fixed to I2S protocol but the output can be changed with the dip switches. The pin 28 has been fixed to ground so the AD189x will be always a master on output. The dip switches and the 10Kohm resistors are not necessary in a final product because you can directly connect the configuration pins to +V or ground. In this case if you see in photo the switch on the position "ON" connect the pin to ground otherwise if you see the switch near the numbers connect is to the pin to +3.3v.
Page 8 of 21 CS4397 To view the bottom of this simple board click here. Follows the measurement on DAC board outputs.
Page 9 of 21 LL7903 con. 1 1000Hz and load 10Kohm (2.4Vrms with thd 0.005%) LL7903 con. 1 100Hz and load 10Kohm In this first configuration the output load should be 10kohm or more. LL7903 con. 3 series 100Hz no load LL7903 con. 3 all sec in series 1000Hz no load (9.57Vrms with thd 0.003%) In this case the load to apply should be 100Kohm (the DAC will see 100k / (8 * 8) = 1500 ohm) but this high output level can be used to drive a buffer like the BU634T or a 2sk216 to create a perfect headphone amplifier. LL7903 con. 3 all sec in series 1000Hz_load DACT 50k (9Vrms with thd 0.004%) LL7903 con. 3 all sec in series 1000Hz_DACT 50k 2SK216 load 1k (9Vrms with thd 0.119%) Follows some measurement of the complete headphone amplifier using the LL7903 as unique voltage stage and BUF634 as current amplifier. This is an original schematic, never seen before, of a h-end no feedback design. Complete headphone amplifier with L7903 & BUF634 withput load (9.5Vrms with thd 0.005%) Complete headphone amplifier with L7903 & BUF634 on 500ohm load (9.43Vrms with thd 0.032%) WM8740 in MONO
Page 10 of 21 To view the bottom of this simple board click here. The WM8740 have internal pull-up or pull-down resistors on configuration pins so are not necessary other external resistors. You can directly connect the configuration pins to +V or ground without use the dip switches. If you see in photo the switch on the position "ON" connect the pin to ground otherwise if you see the switch near the numbers connect this to the pin to +5v.
Page 11 of 21 Many configuration can be set with the 6 x dip switch near the DAC chip.
Page 12 of 21 These DAC chips does not have a current output so in mono configuration (differential) is necessary to add resistors on analog output to merge these. In the schematic there are 2k 1% but I have used a parallel of 2 x 4k 1% to reduce their inductance. Obviously these resistances increase the DAC output impedance so with load less than 10k the output voltage will decrease. In the complete headphone amp. with LL7903 and BUF634 the DACT 50K create an high load but the output level is still good to drive an headphone. In any case in the future I will try to reduce these 2k output resistances. Follows the measurement on this DAC board outputs. LL7903 con. 3 all sec in series 1000Hz_load DACT 50k (3.3Vrms with thd 0.01%) Complete headphone amplifier with L7903 & BUF634 without load (4.4Vrms with thd 0.009%) PCM179x in MONO
Page 13 of 21 To view the bottom of this simple board click here. I have implemented a passive I/V converter with 4 x 100ohm Caddock resistors.
Page 14 of 21 Follows the measurement on DAC board outputs. DAC output board with PCM1794 DAC output board with PCM1794
Page 15 of 21 PCM1794 + LL7903 con. 3 100Hz and load 10Kohm PCM1794 + LL7903 con. 3 1000Hz and load 10Kohm (4Vrms with thd 0.008%) PCM1794 + LL7903 con. 2 1000Hz and load 10Kohm (2.1Vrms with thd 0.025%) PCM1798 + LL7903 con. 2 1000Hz and load 10Kohm (1Vrms with thd 0.05%) AD1955 in MONO As mention in the main differences list with this DAC chip is necessary an active I/V converter and a Microcontroller to set the chip in mono configuration. As microcontroller I have used a Microchip Enhanced Flash type that can be to programmed with a very simple method called: In-Circuit Serial Programming (ICSP ). Using this method is possible program the chip directly on the final board adding only little a connector and a switch to select the 2 modes: run or program. To connect a PC to the ICSP port on the board it is necessary only a Serial Port Programmer available for few money from Spark Fun. I have got some problems with Serial Port Programmer if used with laptop computers so in these case use an USB programmer with ZIF socket like the DIY K128 by kitsrus.com. I have develop the program in C language and I have compile it with a freeware version of the C compiler get from HI-TEC Software in the Download / Demos & free software selection PIC-LITE (Windows), also a Linux version is available. source code hex binary I have used the model 16F876A but also a 12F629 could be good for this porpose. The 16F876A is not supported with the freeware version of the HI-TEC C compiler but can be select the 16F877A without problem because these 2 PIC have the same memory address. The command line to compile the source code is: picl -16F877A AD1955_nospi_v5.c
Page 16 of 21 With the Serial Port Programmer as programmer I have used the freeware Winpic software selecting the "JDM" as interface. The DAC board created for this device is not very simple. To view the bottom of this complex board click here.
Page 17 of 21
Page 18 of 21 I have implemented the I/V converter for the mono configuration with a pair of Analog Device OP275. Follows the measurement on DAC board outputs. DAC output board About passive components types, I don't leave many choices: Caddock resistances on cathode, ELNA Cerafine or Sanyo OS-CON capacitors. A valid alternative to the expensive Caddock is to use a parallel of two normal 1% metallic oxide resistor with inverted phase to compensate some of their inductance. Here is show the connection of the resistors to create the parallel. LL7903 con. 3 100Hz and load 10Kohm (24Vrms with thd 0.017%) LL7903 con. 3 1000Hz and load 10Kohm (24Vrms with thd 0.021%) LL7903 con. 3 1000Hz and load 10Kohm with attenuator set to get 2vrms In this case the max load to apply should be 100Kohm (the DAC will see 100k / (8 * 8) = 1500 ohm) but this high output level can be used to drive a power buffer. LL7903 con. 2 100Hz and load 10Kohm (12Vrms with thd 0.129%) LL7903 con. 2 1000Hz and load 10Kohm (12Vrms with thd 0.021%)
Page 19 of 21 It is possible with a trimmer compensate the operational amplifier dc offset to skip the RC cell on transformer's primary. Follows some measurements with BUF634 or with a 2sk216 connected on transformer's secondary to reduce output impedance and load headphones. The dynamic range could be increase using for the BUF634 a power supply near to +-18V. LL7903 con. 2 BUF634 100Hz attenuator set to max voltage at +-12V (7Vrms with thd 0.09%) LL7903 con. 2 BUF634 1000Hz attenuator set to max voltage at +-12V (7Vrms with thd 0.025%) LL7903 con. 2 BUF634 1000Hz attenuator set to max voltage at +-12 load 100ohm (5.7Vrms with thd 0.256%) LL7903 con. 2 2Sk216 1000Hz attenuator set to max voltage load 1Kohm (11.7Vrms with thd 0.17%) MEASUREMENT RESULTS Follows a new comparing table for these DAC chips about measurement results. Feature Crystal Wolfson Burr- Brown DAC chip Single DAC output [Vrms] Single DAC output THD @ 0dB [%] DAC board + LL7903 connection n. 3 [Vrms] DAC board + LL7903 connection n. 3 [%] DAC board + LL7903 connection n. 2 [Vrms] DAC board + LL7903 connection n. 2 [%] DAC board + LL7903 connection n. 1 [Vrms] DAC board + LL7903 connection n. 1 [%] CS4397 WM8740 mono PCM1798 mono Burr- Brown PCM1794 mono Analog Devices AD1955 mono Burr- Brown PCM1794 mono active I/V 1.132 0.9 0.244 0.480 2.86 4.074 0.006 0.045 0.029 0.038 0.083 0.074 9.57 4.4 4 24 32 0.003 0.009 0.008 0.021 0.059 1 2.1 12.1 0.050 0.025 0.021 2.4 - - 8.32 0.005 - - 0.066 In this table is visible the high distortion reduction after the differential output stage for Analog Device, Wolfson and Burr-Brown that means a very little difference of DAC stage in the same chip. In fact for differential output DAC is not important the distortion level before the differential stage (Single DAC output). SONIC RESULT
Page 20 of 21 TEST1 The system set-up for the first test on September 19, 2006 is the following. One my friend (Pin) with a lot of experience on music reproduction were present at the test. Grades are on a 1-10 scale, where 10 is the absolute best and the test duration has been 3 hours. The shunt regulator has been eliminated after the first 30 minutes because the sound was horrible using this and all the +5V has been generated from a single LT1085. The PCM1798 result is very low but I will make soon a new test using for this chip an active I/V converter like the AD1955. TEST2 The system set-up for the second test on November 1, 2006 is the following. One my friend (Max) (different from the previous test) with a lot of experience on music reproduction were present at the test. Grades are on a 1-10 scale, where 10 is the absolute best and the test duration has been 3 hours. The DAC boards have driven a Lundahl LL7903 connected in 1+1:2 followed by a DACT stepper attenuator. In all the 4 boards all the normal capacitors has been replaced with Sanyo OS-CON. The PCM1798 now use an active I/V converter made using an OP275 (see photo). The amplifier was a prototype of my EL34 PSE with Lundahl transformers, with Jensen copper film interstage capacitors, Svetlana matched quartet output tubes and Jan Philips (12AX7-5814) drivers. One of my best amplifiers. The Loudspeakers cables was the Monitor Cobra and as loudspeakers QUAD ESL 57 from 1972 of one of my friends. TEST3 The system set-up for the second test on November 20, 2006 is the following.
Page 21 of 21 One my friend (Mon) (different from both previous tests) with a lot of experience on music reproduction were present at the test. Grades are on a 1-10 scale, where 10 is the absolute best and the test duration has been 3 hours. The DAC boards have driven a Lundahl LL7903 connected in 1+1:2 followed by a DACT stepper attenuator. In all the 4 boards all the normal capacitors has been replaced with Sanyo OS-CON. The PCM1798 now use an active I/V converter made using an OP275 (see photo). Tthe CS4397 board use 2 x 1Kohm resistors on CMOUT pin like the CDB4397 board and this increase detail. The amplifier was a LM3875 Dual Mono Premium Kit by www.audiosector.com with all Caddock resistances and components close to the Gaincard amplifier by http://www.sakurasystems.com with a price about $,1500 and review on http://www.enjoythemusic.com/magazine/equipment/999/47review.htm. The Interconnect cable was the superlative Tao Analogue Interconnect by Q-Audio. The loudspeakers was my AV system using SEAS COAX driver and project by Punto Musica. Some comments to the result as been given in this case. Feature Crystal Wolfson Burr-Brown Analog Devices DAC chip CS4397 WM8740 mono PCM1798 mono AD1955 mono test 1 5 8.5 4.5 9.5 test 2 6 7 4 9 test 3 6 bit too smooth 8 fantastic 7 too banal 5 too analytic