Integrated Amplifiers

Harman Kardon HK 990 Stereo Integrated Amplifier with Digital Room Correction and Dual Subwoofer Bass Management – Part II


HK 990 Digital Signal Processing and DAC Block

The core of the HK 990 is the digital signal processor (DSP) and DAC that converts the DSP's computations to analog. The function shown in the block diagram directly above (Figure 11) is on the same board as the digital selector block discussed previously.

The switch at the left selects the LPCM data from the digital input board or the analog input board. In the case of the analog input board, the analog signal was converted to LPCM by the Cirrus CS5361 ADC before arriving at the switch.

The DSP in the HK 990 is a TI Aureus TMS320DA708. For room correction, the DSP has two different functions to perform. First, the room must be measured. The DSP generates the test signals to be sent to the speaker, records the resultant acoustic response of the room sent to the microphone (in the room-calibration mode, the microphone signal is routed to the ADC on the analog board), and generates the filter coefficients for room correction. Room-correction systems with on-board DSP can take several minutes to calculate the coefficient because the DSP is not designed to execute these types of off-line computations efficiently. The microcontroller on the main board is also involved in coordinating the room calibration process and transmits information to the MMI board so the user knows where to place the microphones and when it is safe to move them to a new position.

In normal operation, the incoming digital signal passes through the filter bank that is loaded with the room correction coefficients calculated as explained above. A significant part quality of the room correction depends on how well the coefficients are determined during the room measurement process. Systems that do the calculations with an external PC have an advantage.

The other metric of room correction quality is how much filtering can be accomplished in real time by the DSP. Each filter section requires multiplication and addition (subtraction) operations. The faster the DSP operates, more Multiply-Accumulate (MAC) operations can be applied to each sample. MACs (or FLOPS (floating-point operations per second), not the more familiar Million Instruction per second (MIPS), are one of the benchmark for DSP performance. The latter is the typical benchmark for microprocessors

Different DSPs have different MAC ratings that correlate with price. MACs are not the only consideration for a DSP engineer since each DSP has specialized instruction sets and data paths that also affect performance. For example, the DSPs designed for use in AVRs would drain a cell phone battery in a few minutes. Development tools differ significantly for different DSPs and different applications. Without good tools, it may be impossible to complete the design job on time.

In the HK 990, the DSP processor does two other jobs. First, is the bass management function. This will monopolize some of the available MACs. Recall the HK 990 supports two subwoofers that double the MACs for the room correction computations at the subwoofers (calculation requirements for the main channels are unchanged). Second, is to simulate the analog bass and treble control.

The fourth daughter board is a general purpose platform on which the TI Aureus TMS320DA708 DSP chip (Chip package logo shown in Figure 12 above; ™ Texas Instruments) and the associated RAM and Flash memory reside. It is unclear how Harman harnesses the board's computer power in the HK 990. They could have just used the code from the top of the line AVR (AVR 7550HD), leaving most of the computational power of the chip, which would normally process the other unused six channels in AVRs.

Alternatively, the resolution of the room correction system could have been improved over the AVR 7550HD, or the designers could have introduced high-order crossovers to enhance the bass management system. Unfortunately, my measurements of the bass management system indicate it is no better than the ones found on AVRs.

The DSP in an AVR has other chores aside from eight channels of room correction, bass management, and optional jitter suppression (soft ASRC):

      • The decoding of all possible multichannel audio formats entering on the HDMI line to LPCM (example: DTS Master Audio)
      • Volume compression (example: Dolby Volume)
      • Surround synthesis function (example: SRS Circle Surround)
      • Conversion of 5 or 7 input channel inputs to 9 or l1 (example: DTS Neo X).
      • Signal processing to improve the sound of low bit rate compress signal such as MP3
      • THX equalization
      • Virtual Presence Speaker function
      • Hall simulations using delay and reverb (often now expanded for game modes)

To accommodate the myriad functions, two or three DSPs may be present in an AVR. Often two DSP are placed on a silicon die and enclosed in a single package. Much of the code outlined above for an AVR may be proprietary to the DSP vendor. The company designing the AVR picks what options they want and thus have little added DSP coding. The more options, the more MACs used. Low-end AVRs with 5.1 outputs and simple room correction may use one low cost DSP, while a more elaborate unit may employ a trio of DSPs. The DSP vendor profits not only by supplying more DSP hardware, but also via fixed and variable (per unit) fees for the company's code blocks.

AVR manufacturers may develop code for a custom room-correction system, though some DSP suppliers also supply room-correction code. Many Sherwood units and the lower-priced Harman AVRs have room-correction algorithms sourced from Cirrus. Given the substantial research and development effort, an AVR manufacturer hopes to recover these costs in a custom implementation by producing better quality sound when the room correction is selected. In turn, the improved standard of performance should enhance sales.