Test and Measurement: Analyzing computer playback systems
The age of computer-based media storage and playback has brought innumerable conveniences and features to the music-lover. With inexpensive hard drives and configurable media players, your entire music catalog is never more then a few clicks away.
Unfortunately, computer-based playback comes with its own set of unknowns and variables that can be very difficult to analyze and trouble-shoot. Beyond the typical vulnerabilities of computers, problems may exist where you least expect them. For these reasons, Benchmark is providing leading research to determine the limitations of computer-based audio playback systems and the effects of different configurations within these systems. This article discusses the parameters that impact the quality of computer-based audio playback.
When R & D began for the DAC1 USB, Benchmark began investigating the performance of popular computer-based media playback systems. The software and operating systems were objectively tested for performance qualities such as sample-rate and word-length capabilities, volume control performance; bit transparency; bandwidth; and distortion. The most common obstacles for high-quality computer playback are sample-rate conversion, word-length truncation, and various other DSP functions. Results from this research are documented on Benchmark’s ‘Audio Wiki’ (http://www.benchmarkmedia.com/wiki).
Sample-Rate Capabilities
Sample-rate capabilities are a function of the media player, the operating system, and the hardware being used to deliver audio from the computer. Severe distortion may occur if proper considerations are not given to the handling of sample rates. This distortion is a result of poor-quality sample-rate conversion.
The quality of a system’s sample-rate conversion (SRC) services can range from very-high to very-low quality. Very-high quality SRC can be virtually inaudible when up-sampling. When down-sampling, SRC should and usually will apply a low-pass filter at the Nyquist frequency. Therefore, some of the high-frequency information will be lost during a down-sample SRC. Very-low quality SRC will cause significant distortion of the audio. Unfortunately, many computer-based playback systems operate with very-low quality SRC.
Word-Length Capabilities
Word-length considerations for a media player include several factors. It is important to know the maximum possible word length of the connection to the output device. It is also is important to consider how the software will handle longer word lengths. Does the software truncate longer word lengths? Will longer word lengths be dithered to shorter word lengths? Will the entire word length of the digital audio remain unchanged?
Truncating to 16 bits should be avoided when possible as it significantly reduces audio quality. The best media players maintain the entire word length of the digital audio, while other media players truncate 24-bit audio to 16 bits. Certain media players will not even play 24-bit audio files.
The word length of the connection to the audio hardware is equally important. The Benchmark DAC1 USB establishes a full 24-bit USB connection with the computer. Many output devices are limited to 16-bit USB connections to the computer, even if the device has other 24-bit capabilities. Check the manufacturer’s specifications to determine the maximum word length of a device’s computer interface.
Output devices that are limited to 16-bit connections to the computer can often experience distortion due to 16-bit truncation by the playback system. The media player must dither to 16 bits with 16-bit output connections to avoid truncation and maintain low-distortion performance.
Output devices with a 24-bit connection to the computer are much more likely to maintain high-quality performance. In these cases, the media player should ideally implement a 24-bit connection to the output device. Several media players currently available will employ a 24-bit connection to the output device.
Volume-Control Performance
The performance of a digital volume control is directly related to the word-length capabilities of the system. If a volume control truncates to 16 bits, the audio will incur severe distortion. 16-bit volume controls will even degrade 16-bit audio. To maintain high-quality audio playback, 16-bit audio should be played through a 24-bit volume control, followed by a 24-bit data path to a 24-bit output device.
If the volume control dithers its output to 16 bits, truncation distortion will be avoided. However, the noise from the dither may increase the noise heard during playback depending on playback level and the system’s overall signal-to-noise ratio. Dithering to 24 bits is ideal. Truncating to 24 bits is usually acceptable because the resulting artifacts are usually well below the threshold of hearing.
Effects and Other DSP Functions
We do not test non-essential DSP functions, such as ‘sound enhancers’, ‘volume normalizers’, etc. We discourage using these functions, as they generally contradict Benchmark’s goal of truly transparent audio playback.
More Information and Updates
After testing a playback setup, we post all results on Benchmark’s ‘Audio Wiki’ (http://www.benchmarkmedia.com/wiki), which can also be found on the BenchmarkMedia.com website under the ‘Support’ menu. These pages also include recommended setups for particular systems.
It is important to understand that these tests have a ‘shelf-life’. The behavior of computer systems is incredibly dynamic because of the interaction between all software and hardware involved. Furthermore, automatic updates may bring unexpected changes. Without the user’s knowledge, their computer-based audio playback system could have sudden changes in performance!
Each time a media player or operating system is updated, the previous test results and recommended configurations may be obsolete. Do not assume that the newer versions are better then their predecessors, or that any problems with the previous version were corrected. We will try to keep the ‘Audio Wiki’ (http://www.benchmarkmedia.com/wiki) up to date as new versions of software are released.
Technology Update: iTunes 7 for Windows XP and Mac OS X
iTunes has made great improvements with version 7 under both Mac and Windows platforms. The sample-rate conversion, word-length handling, and volume control are all greatly improved. If the correct settings are applied, iTunes 7 can achieve very high quality playback for resolutions up to 24-bit, 96 kHz. This application note discusses the performance and inner workings of the latest version of iTunes (version 7), and also includes specific instructions for achieving the best quality audio playback.
iTunes 7 Performance under Mac and Windows
iTunes sample-rate conversion (SRC) is greatly improved from version 6. In fact, SRC with iTunes 7 is of such quality that it is virtually inaudible (more about this below). iTunes 7 will establish a 24-bit connection with the audio engine of the operating system. iTunes 7 also provides a high-quality, 24-bit dithered volume control – a major improvement when compared to the non-dithered 16-bit controls in prior versions.
The main shortcoming of iTunes is its inability to change sample rates to match the audio it is playing. Ideally, iTunes should automatically change sample rates as needed to match the file being played. In other words, if a 44.1 kHz file is succeeded by a 96 kHz file, the media player should switch from 44.1 kHz to 96 kHz before playing the 96 kHz file. Unfortunately, this is not the case.
Consequently, if iTunes is operating at one sample rate but playing an audio file with a different sample rate, the audio will be sample-rate converted. Although the quality of the sample-rate conversion in iTunes 7 is surprisingly good (and virtually inaudible), it is an unnecessary DSP process.
In contrast, operating system based SRC is usually very poor. The SRC in CoreAudio is very-low quality. Almost equally as poor is the SRC of Windows 2000. The SRC services in Windows XP are somewhat better than CoreAudio and Windows 2000, but are grossly inadequate for high-performance playback. SRC by these operating systems should be avoided whenever possible. The SRC services in Vista seem to be of fairly high quality.
We are currently conducting tests on the latest version of Vista, and will post results soon. We will also be testing the SRC services in OSX 10.5 (Leopard) and will post results soon. Updated test results are listed on Benchmark’s ‘Audio Wiki’ (http://www.benchmarkmedia.com/wiki), which can be found on the BenchmarkMedia.com website under the ‘Support’ menu. Check this site regularly for updates.
Inner-workings of iTunes 7
Under Mac OS X, the ability of all media players to avoid SRC is limited by CoreAudio. This is because the sample rate of CoreAudio will not change without user intervention (done in AudioMIDI Setup). With iTunes v7 operating under Mac OS X, the sample rate of iTunes is set when it is launched, and it will not change unless iTunes is closed and reopened. iTunes v7 determines its sample rate by matching it to the sample rate at which CoreAudio is operating at the time of iTunes’ launch. Once the sample rate of iTunes is set, it will only operate at this fixed rate until iTunes is reopened, even if the sample rate of Core Audio has changed. When iTunes is reopened, it will, once again, set its sample rate to match the current sample rate of CoreAudio. This may seem awkward (and it is), but it seems to be a work-around to avoid CoreAudio’s SRC services.
iTunes 7 provides its own very-high quality SRC as an attempt to avoid the very-low quality SRC of CoreAudio. iTunes will convert all other sample rates to the sample rate at which iTunes is set. If the sample rate of CoreAudio is not changed, it will remain matched to that of iTunes. In that case, CoreAudio will not apply SRC to the audio.
However, if the sample rate of CoreAudio is changed (in AudioMIDI) while iTunes is open, CoreAudio will then convert the sample rate of the audio it is receiving from iTunes. For example, if CoreAudio is operating at 96 kHz when iTunes is launched, iTunes will solely operate at 96 kHz. If CoreAudio is then changed to 44.1 kHz, all the 96 kHz audio coming from iTunes will be converted to 44.1 kHz by CoreAudio’s SRC. Furthermore, if iTunes is playing a 44.1 kHz audio file in this case, iTunes will convert it to 96 kHz, then CoreAudio will convert it to 44.1 kHz. In this case, SRC is being applied twice.
Under Windows operating systems, software applications have the ability to change sample rates without user intervention. Unfortunately, iTunes does not take advantage of this capability. Under all Windows operating systems, iTunes will only play at a fixed sample rate (specified in the ‘QuickTime Preferences’ control panel found in QuickTime). This is an unnecessary limitation that can have some impact on quality if SRC is evoked. It is even possible to set the sample rate of iTunes (QuickTime) so that it neither matched the file sample rate nor a supported sample rate of the output device. In this case, both iTunes and Windows’ Kmixer would employ SRC.
Fortunately, as long as the sample rate of iTunes is within the capabilities of the output device, it is highly unlikely that the SRC services of Windows XP and 2000 will be invoked. Under these operating systems, SRC is only invoked if multiple audio apps are streaming audio at different sample rates to Kmixer simultaneously. More information about this can be found on our ‘Audio Wiki’ (http://www.benchmarkmedia.com/wiki).
Similar to iTunes v7under Mac OS X, changing the sample rate of iTunes v7 under Windows requires the user to restart iTunes after the preferred sample rate is set in QuickTimes. Consequently, iTunes will invoke SRC when playing any audio files with sample rates different from the rate at which iTunes is fixed. Although the SRC within iTunes is of very-high quality, it is an unnecessary DSP. Furthermore, if the sample rate of the audio file is higher then the rate at which iTunes is fixed, the audio will be down-sampled. In this case, the audio will suffer a reduction of analog bandwidth.
Changes Since iTunes v6
iTunes v6 under Mac OS X would always operate at the sample rate of CoreAudio. When the user changed the sample rate of CoreAudio, iTunes would immediately change its sample rate to match CoreAudio, even if the change was made when iTunes was open. However, CoreAudio would not automatically adjust sample rates to match the file being played. This made audio playback extremely susceptible to SRC. Consequently, audio playback with iTunes v6 under Mac OS X suffered from the SRC problems of CoreAudio when the sample rate of CoreAudio did not match the sample rate of the audio file being played.
Under all Windows platforms, iTunes v6 suffered from SRC problems because it failed to change its output sample rate to match the sample rate of the file being played. Although this hasn’t changed in version 7, the SRC in iTunes v6 was of very-low quality. Consequently, incorrect sample-rate settings resulted in significant distortion under iTunes v6. Like iTunes v7, It was possible to set the sample rate of iTunes (QuickTime) so that it neither matched the file sample rate nor a supported sample rate of the output device.
Under both Mac and Windows operating systems, the volume control in iTunes v6 induced severe distortion due to 16-bit truncation. The only saving grace about iTunes v6 was that if everything was set exactly right and if all files have the same sample rate, iTunes v6 could provide bit-transparent 16-bit playback under both Mac and Windows platforms. This is too many “if’s”!
Recommended Setups
Optimal settings for iTunes differ depending on the capabilities of the audio output device being used. The following recommendations are for the Benchmark DAC1 USB and other devices whose computer interfaces are capable of 96-kHz, 24-bit operation.
The Volume Control in iTunes v7 is very-high quality under both operating system. It is very well designed and operates at 24-bits for audio devices that support 24-bit operation. The end user should not hesitate to use the volume control in iTunes 7.x under either operating system.
For users of iTunes under Mac OS X with 96/24 output devices, we offer the following two different, recommended solutions:
- The ‘Set It And Forget It’ solution for iTunes 7.x: Before opening iTunes, set the sample rate of CoreAudio (in AudioMIDI Setup) to 96 kHz. Do not change the sample rate of CoreAudio unless iTunes is restarted after the change is made. This solution will prevent CoreAudio from applying SRC, as the quality of CoreAudio’s SRC is horrible. Also, by having iTunes locked at 96 kHz, all audio with sample rates below 96 kHz will be up-sampled to 96 kHz. This will cause virtually no loss in sonic quality, as the quality of iTunes’ SRC is very good – virtually inaudible. Also, by avoiding down-sampling by iTunes, this setting will never result in a loss of bandwidth (except with sample rates greater then 96 kHz).
- The ‘Bit-Transparency For Each Sample Rate’ solution: *CAUTION: This solution is rather cumbersome, offers virtually no quality improvement over the first solution, and can easily be mis-configured which can cause severe distortion. Before opening iTunes, set the sample rate of CoreAudio (in AudioMIDI Setup) to that of the audio you will be playing. Do not change the sample rate of CoreAudio unless iTunes is restarted after the change is made. This solution will prevent CoreAudio from applying SRC, and avoid SRC by iTunes for all audio with the same sample rate at which iTunes is fixed.
For users of iTunes under Windows XP and 2000 with 96/24 output devices, we offer the following two different, recommended solutions:
- The ‘Set It And Forget It’ solution for iTunes 7.x: Before opening iTunes, set the sample rate of QuickTime to 96 kHz. That is all. By having iTunes fixed at 96 kHz, all audio with sample rates below 96 kHz will be up-sampled to 96 kHz. As the quality of iTunes’ 7.x SRC is very good (virtually inaudible), this will cause virtually no loss in sonic quality. Also, by avoiding down-sampling by iTunes, this setting will never result in a loss of bandwidth (except with sample rates greater then 96 kHz).
- The ‘Bit-Transparency For Each Sample Rate’ solution: *CAUTION:This solution is rather cumbersome, offers virtually no quality improvement over the first solution, and can easily be mis-configured which can result in a reduction of bandwidth. Before opening iTunes, set the sample rate of QuickTime to that of the audio you will be playing. You must change the sample rate of QuickTime first, then restart iTunes. This solution will avoid all SRC as long as all audio files match the sample rate at which iTunes is fixed.
Further Updates
As always, check for new test results and updates on Benchmark’s “Audio Wiki” (http://www.benchmarkmedia.com/wiki), which can be found on the BenchmarkMedia.com website under the ‘Support’ menu.