Signal Analysis of THD+N with dB3000S

Introduction  
 
Lavry Engineering's latest revision to the Model 3000S Sample-Rate Data Format Converter consists of THD+N (total harmonic distortion and noise) measurement capabilities. The new function expands the usefulness of this "digital audio all in one tool". The new feature offers the high performance and ease of use found in all the other features (data format conversion, sample rate conversion, Acoustic Bit Correction, reference meter bridge and test tone generation). 24 bit tone generation and THD+N measurement accuracy in excess of 122ÝdB provide sufficient margin for the next generation of digital audio equipment.  

The Model 3000S's test tone generation and Signal analysis are independent of each other. This flexibility allows performing either function separately or both simultaneously. Distortion and noise  

Real hardware generated tones contain both distortion and noise. Harmonic distortion is referred to as energy residing at multiples of the tone frequency. Noise is all other undesirable energy.  

Some common sources of distortion:  

* Amplifiers: non linear signal processing causes distortion. While mostly corrected for by use of negative feedback, amplifiers tend to degrade at higher frequencies.  

* Component imperfection: distortions occur when component values depend on the signal. For example, capacitors tend to counteract a changing signal (dielectric absorption).  

* Digital truncation: limitations of word length (number of bits) is a non linear process. Such distortions increase at lower signal levels.  

By definition, noise is not all random (tones occurring at non harmonic frequencies are considered noise). Well known and understood is the critical noise requirement associated with signal amplification. Often overlooked is the accumulated noise due to connecting of many units in series.  

Common sources of random noise:  

* Resistor noise (flat frequency distribution) increases for larger value of resistance.  

* Semiconductor noise, mostly flat frequency distribution. Increased noise levels at very low frequencies usually occurs below audible frequencies.  

* Capacitor noise, inconsequential for higher values, has become a performance limiting factor with the introduction of very small capacitor values incorporated in modern semiconductors such as oversampling sigma delta converters.  

* Digital truncation: limitations of word length (number of bits) in A/D converters, signal processors and more. The problem grows with increased amount of processing. Common sources of non-random noise:  

* Coupling to analog signal path: AC power line, RFI/EMI, coupling of digital signals to analog path, inadequate power supply rails and more.  

* Digital truncation: limitations of word length (number of bits) in A/D converters, signal processors and more. The problem increases for low level signals.  

* Limit cycles: cyclical patterns behavior in feedback based digital signal processing (such as sigma delta converters and IIR filter structures).  

Listening tests and measurement of individual equipment in the audio chain does not guarantee optimum THD+N performance. Setup optimization is a very complex subject. Top recording engineers blend artistic considerations and engineering know-how into the process. The following discussion does not deal with artistic aspects. We acknowledge the great importance of artistry in music production, but are bound to limit our discussion to measurable and objective phenomena.  

Optimizing THD+N (some engineering considerations):  

Analog amplification: good signal to noise ratio requires "early" signal amplification, but with careful attention to tradeoffs between distortions and noise.  

Analog attenuation: undesirable from noise standpoint, may be required to accommodate signal range limitations of various gear.  

Jitter: clock jitter in A/D, D/A and Sample Rate Converters degrades THD+N. Reference D/A clock jitter in a studio may have little to do with the end product quality, but may make the monitoring process difficult. Sample rate converters perform best with low jitter on both incoming and outgoing clocks.  

Configuring proper digital chain: whenever unit A may drive unit B or visa versa. A good "rule of thumb" is to have the better performer drive the lower performer. A quality digital device utilizing 24 bit words is limited to 16 bits when driven by a 16 bit device. The compounded outcome is that of "2 x 16 bit devices". Reversing the order allows the first process to retain its high accuracy, leaving a compounded outcome of one 16 bit device.  

Measuring THD+N  

The common method for measuring THD+N is based on feeding a "device under test" with a quality reference test tone and measuring the undesirable energy (THD+N) at its output. Lavry Engineering's Model 3000S provides the user with a reference test tone. The processed tone (or any other source) may be fed back to Model 3000S input. The input signal is filtered by a very sharp notch to separate the desired signal component from the undesirable energy (THD+N). The undesirable energy (THD+N) is then displayed inÝdB (referenced to full scale).  

The notch filter must be very deep and narrow. Notch depth assures that no energy at the fundamental frequency "leaks" to the meter. Narrow notch is required to leave noise and harmonics intact. How steep should the notch be? For theoretical tones one may strive for the steepest notch possible. Real applications require full attenuation over a slightly wider frequency range to accommodate possible small jitter of both sampling clock and the tone itself. Model 3000S provides sufficient attenuation over about .1% (allowing about 20 nsec jitter).  

Additional 20 Hz high pass filtering ensures that very low frequency components (such as DC and other low frequency inaudible energy) does not alter the outcome.  

Model 3000S begins its measurement by locking to the test tone frequency. There are 2 modes (user selectable) for achieving lock:  

* Auto mode: Proper locking in the presence of noise and distortion requires "reasonable" signal to noise and distortion ratio. When using Auto mode we recommend starting the test with a large enough signal (full scale signals are ideal, but locking will take place at 40ÝdB over the noise and distortions floor). Pressing the "Go" button sets the notch in place, allowing THD+N measurements at any level.  

* Normal mode: when using Model 3000S tone generator, the notch frequency follows the tone frequency settings. Normal mode frees the lock mechanism from any signal to noise restrictions. This mode requires the tone sample rate and incoming signal sample rate to match within +/- .1% of each other (testing Sample rate converters requires Auto mode).