Thoughts about Music Reproduction

If you have not experienced REALLY good music reproduction on a high-end system, I urge you to go to a specialty store (hopefully an independent dealer, but they are sometimes not conveniently located). Usually, they will have a good selection of music, but you can bring your own CDs or LPs if you have favorites. Ask questions about different speakers, amplifiers, and reproduction devices. And listen. If you do this, you MAY get hooked. At that point, you can decide if you want to invest in off-the-shelf components or go down the DIY path.

Of course, a lot depends on your musical tastes. If you lean toward heavy-bass hip-hop or electronica, you can take some shortcuts. But if you are into well-recorded acoustic music, or even classic rock, folk, or jazz, you will probably be a bit more discerning about what comes out of your transducers. I think this is especially true for musicians.

Here is my list of the critical components in an audio chain (in order):

• Speakers (or headphones if you prefer) - the speaker's job is to move air in response to the music signal. All speakers introduce some distortion in the music signal. The idea is to minimize this and there have been endless numbers of attempts to do so. A secondary characteristic is frequency response. Traditionally, speaker designers tried to make them "neutral" so the response curve was as flat as possible, which is very difficult to achieve. [On the other hand, Linkwitz believed that speakers had to be deliberately made somewhat non-flat in order to correspond to the way we perceive music. See more on his website].

• The music itself - this will typically be a recording of some sort, on storage formats like LP, tape, CD, DVD, streaming from the internet, or local storage devices. The quality of recordings varies all over the map, as well as the methods used to encode them onto the media.

• The music conversion device - This can be a turntable, a tape deck, a CD player, a DVD player, an audio streamer. A digital device will require a D/A converter before amplification (except in the case of all-digital amplifiers which are not common).

• The amplifier - an amplifier's job is deceptively simple: convert a low-voltage signal into a high-level signal capable of driving speakers without introducing undesirable artifacts. This means both voltage gain and current driving ability into low impedance loads.

Amplification

You may notice I listed amplifiers last - that is because they are perhaps the easiest to get "right". But they can muck up your listening experience if they are not right. The amplifier should disappear sonically, but this is not so easy to achieve. Amplifiers have to be designed to work with a wide variety of speaker loads, so designers use a one size fits all approach, which can result in compromises. Also, manufacturers love bragging rights and will tout huge power numbers and astonishingly low distortion figures. This is usually marketing hype and may not reflect what an amp actually sounds like playing music.

Of course, there has been a long-running debate among the Objectivists (those who, at one extreme, only believe in measurements) and Subjectivists (those at the other extreme who will say that listening is the only important criteria). Measurements are great and can lead to amplifier designs with vanishingly small amounts of that which can be measured. A pure objectivist will claim that is all that matters and will make the case with data. The subjectivists will say there are things that can be heard but which cannot be measured. The problem that subjectivists have is describing this with terms similar to what wine enthusiasts use to describe different tastes in wine. Language is completely inadequate when it comes to describing differences in taste, smell, or sound, not to mention individual differences in perception. Audio marketing is a combination of the two but reviewers tend toward the subjective and things start to get silly. What is important is how an amplifier sounds with your speakers and your music, and how close you think it is to live performance, free of any artifacts that did not exist in the original.

You may be familiar with the different "classes" of amplifiers: A, B, A/B, D, G, H, variants thereof, and so on. They all have different characteristics and it is not possible to say which is best for you. In general, class A amps are often regarded as the most "musical" but least efficient (meaning they dissipate a lot of energy even when doing nothing). They are typically low-power output devices. To say that class A amps are the most musical is a broad generalization, of course, since it is entirely possible to make a horrible-sounding class A amp that is easily outdone by any of the others. Personally, I prefer class A (assuming it is well done) where it makes the most difference - in the mid-to-high end of the audio range with efficient speakers (which is why I built this amp). For the low end, like woofers, I think A/B is fine (which is why I built this amp) and for subwoofers, I think a class D is just fine. A carefully designed class A amp can sound glorious, whether built with tubes or solid-state devices. They can be either single-ended (meaning a single output device) or push-pull. Class A amps completely avoid what is called cross-over distortion, common in class B amps. That occurs when one output device shuts off before the complementary device can begin to conduct during the output waveform. In the early days of solid-state amps, this was a common flaw and lead to the label of "transistor sound" which was bad. Real bad.

See my article on Tubes vs Solid State for thoughts about the two technologies.

I have lately become interested in Class D for things other than subwoofers, so I might start looking into building a DIY amp using Class D modules if time and budget allow. See my musings on class D in a separate section.

See this article at audioholics.com for a succinct summary of audio amplifier classes..

Distortion

So what about distortion and measurements? There are various types, the most common of which is Total Harmonic Distortion, or THD. Almost all amplifiers will have THD specified. For power amps, the spec is usually for a given output power at its maximum rated level. THD is measured using a single pure sine wave typically at 1 kHz. Hypothetically, if the amp reproduces the sine wave perfectly, it has zero THD. Practically, all amplifiers will exhibit some THD, even if the number is small. Noise measurements are often combined with THD to produce a THD + Noise figure (these days referred to as Signal-to-Noise And Distortion or SINAD), expressed as a percent or decibel level. THD is a gross measure, since it represents the sum of all harmonics present in the output, regardless of their numerical relationship to the fundamental. Generally, even harmonics are less objectionable than odd harmonics, so THD is a bit disconnected from what is pleasing to the ear and can be misleading. 

Transient Intermodulation Distortion, or TIM, was identified in the early 1970s as being caused by amplifiers with low slew rates (also mentioned in the literature as Slew Induced Distortion or SID); that is, a lag in the time it takes for an amp to ramp up to the full value of a step-function signal. This was prominent in early small-signal op-amps when they were first used in audio with not great audible results, even though 1 kHz THD measurements we fine. Tube amps were less prone to the phenomenon since tubes are inherently high bandwidth devices. This spec is rarely mentioned by manufacturers since it has largely become a moot point for well-designed amps. The key to minimizing TIM is to design the amp with a bandwidth of at least 100 kHz. This is fairly easily done in class A, B, and A/B amps, but may not be so easy with class D, since they have heavy ultrasonic filtering. Excessive negative feedback, used to correct badly non-linear circuits, is also thought to contribute to TIM (the key, obviously is to design circuits with open-loop behavior as linear as possible, then apply small amounts of negative feedback). Unfortunately, no standards exist for measuring this type of distortion, even though it has been shown to have an audible effect.

Other Amplifier Misbehaviours

• Damping Factor DF) - This is a measurement of the amplifier's ability to control a loudspeaker load. A typical moving coil speaker when energized and then unenergized, becomes a generator and produces a back EMF. If uncontrolled, the speaker will oscillate and produce unwanted sounds. In order to control this, especially with low-frequency drivers, the amp must have a good damping factor and low output impedance. People will say that some amps do not have a "tight" bass sound. This is usually due to a low DF. Tube amps typically did not have high damping factors due to the output transformer. Speaker cables with high impedance (long runs with small gauge wire) will swamp out any effect, but it is generally considered "good enough" for an amp to have a DF of 50 or greater. Also, speakers vary widely in terms of how they respond to amplifiers. Above 200, there is little to be gained. My Class A amp described here has an adjustment for DF and can be tuned for a particular speaker load.

• Clipping - an amp with limited power output swing (voltage and/or current) can exhibit a saturation effect on loud passages or transients. Usually not a problem with quiet music and efficient speakers, but can be horrible sounding otherwise. The only solution - if you like loud - is to have a very over-specced amp and efficient speakers. But note that speakers can also be overpowered and exhibit a "bottoming out" effect once they exceed their rating. Usually, the louder they are played, the more distortion they will exhibit.

• Phase response - amplifiers can exhibit phase shifts in the signal at different frequencies. This is a very subtle effect and is almost never published. The auditory effects of phase shift are debatable, but ideally, there should be little or none. High bandwidth amps (up to 1 MHz) will generally show negligible phase shifts in the audible band.

• Crosstalk - applicable only to stereo amps, this usually occurs due to shared power supplies (as opposed to monoblock designs ) between channels or poor wiring practice. Crosstalk results in a smearing of the sound across channels, reducing the perceived placement of instruments in the listening space.

Musicality of Amplifiers

There is actually a technical explanation for some class A amps potentially sounding more "musical". This has to do with the fact that a well-executed class A design will emphasize the second harmonic distortion of fundamental frequencies over odd-order harmonics. Second (and other even-order harmonics) are more pleasing musically. Consider a vibrating string in a musical instrument. The fundamental tone is the frequency at which the string vibrates. But the string will also have overtones, which are harmonics (multiples of) the fundamental. These overtones, along with the construction of the rest of the instrument, are what give the instrument its timbre. Even harmonics predominate in stringed instruments and a small amount of additional even harmonics added by an amplifier are almost impossible to hear. They result in only a very slight perceived change in the timbre of the instrument. Odd-order harmonics, on the other hand, can create a sort of dissonance that could make the sound harsh or unnaturally bright. Another subtlety is the phase of the second harmonic relative to the fundamental created by the amplifier. Negative phase second harmonics are often thought to be more pleasing to listeners. The design of the amp and the characteristics of its components can influence the harmonic phase. Nelson Pass wrote this article which discusses the importance.

I am probably not explaining the importance and subtlety of this harmonic topic very well, so I will defer to a seminal work by Sir James Jeans, published in 1937(!), titled Science and Music, available from Dover Press. This is from a section titled "Vibrations of Strings and Harmonics", starting on page 86:

"The second harmonic adds clearness and brilliance but nothing else, it being a general principle that the addition of the octave can introduce no difference of timbre or characteristic musical quality. When the second harmonic is of equal strength with the first, it produces much the same effect as adding the octave-coupler on an organ or harmonium or playing in octaves, instead of single notes, on the piano.

The third harmonic again adds a certain amount of brilliance because of its high pitch, but it also introduces a difference of timbre, thickening the tone, and adding to it a certain hollow, throaty or nasal quality, which we may recognise as one of the main ingredients of clarinet tone 

The fourth harmonic, being two octaves above the fundamental, adds yet more brilliance, and perhaps even shrillness, but nothing more, for the reason already explained.

The fifth harmonic, apart from adding yet more brilliance, adds a rich, somewhat horn-like quality to the tone, while the sixth adds a delicate shrillness of nasal quality. As the table on p. 73 shews, all these six harmonics form parts of the common chord of the fundamental note, and so are concordant with this note and with one another. 

The seventh harmonic, however, introduces an element of discord; if the fundamental note is c', its pitch is approximately bb'", which forms a dissonance with c. The same is true of the ninth, eleventh, thirteenth, and all higher odd-numbered harmonics; these add dissonance as well as shrillness to the fundamental tone, and so introduce a roughness or harshness into the composite sound. The resultant quality of tone is often described as "metallic", since a piece of metal, when struck, emits a sound which is rich in discordant high tones."

Interesting! So if we translate this analysis to amplifiers (or any other audio reproduction electronics), we can clearly see that in order to have the highest possible fidelity and musicality, we need to minimize the addition of harmonics, particularly the higher-order ones. Each spurious harmonic introduced by the electronics will add coloration to the sound and this could well explain why critical listeners hear differences in equipment. 

So you can see that these objective measurements of amplifiers are not exactly simple, and that has given rise to even more debate among the subjectivists.

Preamps

Do you need a preamp? Preamps are referred to as line-level components - usually involving signal levels 2 Volts RMS or less. It can be passive (meaning no additional signal amplification) if your source devices produce sufficient voltage to drive your power amp. You will need a way to attenuate (as opposed to boosting) the signal to control the volume. Also, you will want a way to select from several sources. All these can be handled by a preamp, but it does not necessarily need to provide gain. See my project page for my implementation. Some preamps feature simple buffer stages with no gain, while most active preamps have selectable gain amounts. In either case, you have active devices in the signal path, so additional distortions/musical colorations will occur. I think if either of these features exist in the preamp, they should ideally be discreet class-A topologies. However, there are some excellent opamps available today that can rival discreet designs, like the TI OPA1656 which many DIY audiophiles are using to build line-level circuits.

All-in-one or integrated amplifiers combine the power amp with a preamp and usually an input selection mechanism. Some people prefer these since they are (theoretically) matched and more convenient, taking up less space. But they also provide less flexibility in terms of being able to swap out just the amp or preamp.

Receivers are much more common because the industry has had to address the audio-visual (A/V) market; that's where the money is. These include tuners, HDMI video switching, and multi-channel surround sound. Some receivers can be very good at audio, but again there are usually compromises to be made in terms of quality. The really good ones end up costing as much or more as separate components. All A/V receivers have digital signal processing (DSP) capabilities for surround effects. In order to get "pure" audio reproduction, you need to shut off or bypass the DSP circuitry. Most receivers have a stereo mode to allow this. If you want to combine hi-end audio with your A/V setup, you can usually find a reasonably priced receiver and then opt for separates for the main stereo listening experience. That's the way I would go if I were to build a combo A/V and audio listening system.

Music Conversion

D/A Conversion and Compression

Streaming has taken us into a whole new realm, and new digital formats have become common. Initially, all we had was MP3 (mainly for portable players like the iPod), a lossy encoding format that audiophiles avoid. Very high bitrate MP3 can sound better, but the stream is never "bit perfect" due to lossy compression. Lossy compression made sense back when storage and bandwidth were very expensive, but these days it does not. There simply is no legitimate reason to do lossy compressed audio anymore. In fact, bit rates and resolution (measured in terms of bits) have been increasing. 24-bit 96 kHz (and higher) encoded audio is now commonly available. Note: Only the very best D/A professional-level converters can even approach 24 bits. The very best consumer DACs run about 21-22 bits of dynamic range. This is around 128 dB. The so-called limit of audibility is about 115 dB (down from 0 dB) as a reference. But even converters that can only achieve 18-20 bits of resolution (even though they are advertised as 24-bit) will likely sound better than 16-bit CD quality. 16-bit audio has a dynamic range of 96 dB and must be noise-shaped using dither of the LSB to achieve greater than that. This is a bit of a bandaid and is unnecessary with greater bit depths. 

There are two considerations here: bit depth and sample rate. For an interesting discussion of the historical reason for de facto industry standards, see this article. It explains why we have sampling rates of 44.1 kHz and its multiples, 88.2 and 176.4 vs. 48 kHz and its multiples, 96 and 192. Since the article is about what people in the industry use (users of Pro Tools recording software) there are some interesting comments. One commenter says he is fine using 44.1 for rock but uses 96 for chamber and classical because it sounds better in preserving the subtleties of the acoustic instruments.

I saw a YouTube video recently "proving" that all you need is 16 bits and 44.1 kHz because the output filter in the DAC used was able to convert an ugly staircase signal into a pretty sine wave as viewed on an oscilloscope. Of course - if the original signal is a perfect sine wave with no harmonic components (zero distortion) it will be perfectly reconstructed. Plus, a "brick wall" filter can convert almost any waveform of a fixed frequency into a perfect sine wave. But that means the original signal was corrupted prior to conversion due to inadequate sampling. Another problem is that people misinterpret the Nyquist-Shannon sampling theorem when it is applied to real-world signals and problems with band limiting the signal before conversion. You are throwing away information by filtering the input signal prior to conversion (which is a prerequisite for the Nyquist theorem to hold), especially transients - which are critical in music.

Similarly, there are pundits who will argue that you will never need more than 16 bits and will cite psychoacoustic studies "proving" that people cannot even perceive the difference between 8 bits and 16 bits. This is bunk. I can tell you that all else being equal, there is a clear difference between 16-bit and greater bit depth and sampling rates above 44.1 kHz. Trained listeners (e.g. professional musicians) have demonstrated the ability to perceive differences in double-blind studies. Now the "all else being equal" caveat is a big one, because if you just take a lousy original recording, perhaps one that has been heavily compressed (little dynamic range) or stored on noisy analog tape and then re-mastered in hi-rez, what have you gained? Probably not a lot. Imagine taking an old 78 RPM record and converting it to 24-bit, 192 kHz format - will it sound better? Nope - but at least the imperfections in the 78 recordings will be very accurately reproduced.

The audio recording industry has largely moved to 24-bit 96 kHz recording as a standard, and all D/A manufacturers. support it. The idea is to throw away as little as possible in the audio chain. So if the audio chain is hi-rez from start to finish, then there is a clear benefit. If a studio has made recordings using an end-to-end hi-rez signal chain, why would you want to decimate that down to 16-bit 44.1 KHz? Of course, there are a lot of "remastered" hi-rez recordings made from poor quality masters and mostly do not sound any better than 16-bit.

Back on the subject of compression, none of this means you can't have lossless compression formats. These don't have as small a file size as MP3, but they still save on storage and bandwidth. Popular formats are FLAC (Free Lossless Audio Codec), ALAC (Apple's format), WMA Lossless (Microsoft's), and Ogg, among others. FLAC has become very popular since it is non-proprietary and can encode various bitrate streams, ranging from 16-bit 44.1 kHz (CD) up to 24-bit 192 kHz. Uncompressed formats such as WAV (Waveform Audio Format - developed by IBM and Microsoft - result in much larger file sizes than FLAC, so there is not much benefit in using it for music listening, except for manipulating sound recording. If you are familiar with DAWs (Digital Audio Workstations) used in the recording industry, they typically can manipulate WAV files easily.

The most common encoding method is multi-bit PCM (Pulse Code Modulation) and is supported by all the above formats.

DSD (Direct Stream Digital) is different. DSD was originally created for SACD, then faded away, and is now back. DSD is a single-bit format that uses a very high-frequency carrier rates (2.8 MHz, 5.6 MHz, and even 11.2 MHz). Unfortunately, it is harder to manipulate in the digital realm than PCM, but many audiophiles claim to like the sound of DSD better than PCM. It is not compatible with FLAC and requires a different decoding technique. Unless you have very high-end D/A converters, the differences are small between DSD and high-resolution PCM. Most of the better converters these days can handle both, so you can choose your music format and decide which sounds best in your system.

To make things even more confusing, we have DoP (DSD over PCM). In this case, PCM is used as a carrier for the DSD stream, making it compatible with conventional 24/192 PCM transmission channels like USB (Universal Serial Bus, common on all computers), AES (Audio Engineering Society standard interface), and S/PDIF (Sony/Philips Digital Interface - the consumer version of AES).

And finally, I should mention MQA, Master Quality Authenticated. This is a new format that purports to capture all the nuances of original master recordings and requires special hardware (or software on a computer) to take full advantage. Several services in the list below are beginning to offer MQA-encoded music. MQA has resulted in a lot of controversy, ranging from "snake oil" comments to the best thing to happen to audio. So far, I am leaning toward the snake oil end of opinions - do we really need yet another encoding "standard"? BTW - it is proprietary, so not exactly a standard in my book. Just recently (April 2023), MQA filed for bankruptcy protection, so it will be interesting to see what the future holds. Tidal is the largest provider of MQA-encoded streams, so they may take it over.

Update - July 1st, 2023 - According to this article, Tidal has announced that FLAC is now their preferred encoding method for hi-rez streaming, displacing MQA. So while they apparently will continue to support MQA, there is little doubt it is on the way out.

My suggestion is to not worry about the technical differences too much, and just make or buy the newest, highest resolution D/A converter you can afford. DACs have continually improved recently due to competition among chip vendors like ESS, AKM, and Cirrus. This has allowed DAC manufacturers to play the resolution and THD numbers game cost-effectively.

See my latest music streamer project for an example of a very good D/A converter for DIY that costs little and has excellent performance.

The are several online sources of hi-definition audio these days:

• HDtracks - one of the original hi-resolution download services started by David and Norman Chesky of Chesky Records fame

• Acoustic Sounds - unfortunately, this company closed down at the end of 2020, citing the rise of streaming and regulations

• Deezer HiFi - appears to be CD-quality only, but I include them anyway since things are always changing

• HiRezAudio - offer both streaming and downloads in multiple formats

• ITrax - claims to be the original hi-rez download site. All tracks mastered in 24bit/96 kHz.

• Primephonic - appears to be streaming only, focused on classical

• Qboz - lots of hi-rez, both streaming and downloads

• TIDAL - 70 million streaming tracks

• Linn Records - Offers mainly jazz and classical downloads - "Studio Master" quality lossless files, mostly 24-bit and up to 192 kHz

• Amazon Music HD - a mix of CD-quality and hi-rez streaming. Downloads appear to be limited to MP3 for now.

The Music Itself

I won't bother pontificating much on the music itself, since personal tastes vary so much, but it is worth touching on the care that goes into capturing the music in the first place. This usually comes down to the producer (and sometimes the artist), the recording setup used, and the recording engineer(s). State-of-the-art recording studios today use the very best equipment in their signal chains. As mentioned, DAWs keep digital domain audio in high-resolution lossless formats, so little is lost in terms of audio information. Usually, with multiple microphone setups and multi-tracking, all the separate tracks are stored individually. Then it is up to the mastering engineer to pull these into a stereo format. This is where things can go off the rails, since the mastering person may be creating the mix the way he or she thinks best, not necessarily what the artist intended or what an audience would have experienced had the artist performed live. Another issue can arise with the type of speakers used to listen to the master re-mix. Since speakers vary greatly, and most commercial studios cater to the more popular forms of entertainment (less so acoustic music), there can be a mismatch with what an audiophile expects to hear. This is why you hear so much variability in recordings. One thing you can do is find something that you really enjoy listening to and seek out other recordings from the same producer/engineering team. This might require some research, e.g. looking up the discography of a particular person.

I was encouraged recently when I read that Jack Vad, the sound engineer for the San Francisco Symphony, uses Linkwitz-designed speakers for the final master recordings of the symphony. See his quote on this page.

Speakers

Finally, speakers. This could be a whole encyclopedia! I'll start a whole new article soon with my opinions.