Modems: Run It on the Host

Moore’s Law states that the speed of a general-purpose processor doubles

every 18 months.

While the processor manufacturers compete with each other to meet this

challenge, PC manufacturers face an interesting problem. How do they convince

the end customer to buy a PC with a much higher speed? After all, if the needs

are limited to Internet access, the 1 GHz machines are more than enough.

The only way to upgrade the customer is by offering applications that require

higher processor bandwidth.

In order to attract new customers to buy high-end PCs and motivate the

existing customers to upgrade, PC manufacturers are forced to bundle more and

more applications at attractive price points.

PCs as Host Systems



Modems, for instance, typically require a very strong DSP or need hard-wired

signal processing ASIC blocks. With the significant advances happening in

semiconductor technology, it is now possible to have highly cost-effective

systems-on-chip (SoCs) offering very complex functionalities. Typically, the

system that needed an entire board with discrete components five years ago, can

now be integrated onto a single chip. Typically, the power consumed by the chip

and the chip cost are the significant differentiators.

The other approach that’s now gaining momentum is host computing. This

approach involves running the applications in a soft form on the CPU itself. Of

course, some minimal hardware support may be unavoidable. The approach relies on

the fact that most of the bandwidth on a general-purpose processor used in a PC

remains unused most of the time. If a PC is typically used for Internet

browsing, e-mails and documentation, more than 50 percent of the CPU bandwidth

should be idle on a 1GHz CPU. The amount of idle bandwidth only increases as the

processor speed increases.

Gone are the days when one would buy a bulky modem card along with the PC.

The modem cards are much smaller and sleeker now. For MIPS-intensive modems,

such as the 33.4 kbps V.34, the typical approach earlier was to implement the

entire data pump on a platform with DSP and RISC cores, complemented by hardware

accelerator blocks. Only the higher layers would run on the PC. The typical

approach now is to avoid any DSP or RISC processor and hardware accelerators

completely, and run the entire modem on the host CPU itself. Of course, the

analog circuitry needed to interface to the telephone line needs to be there in

either case.

Cost Benefit for Modem Manufacturers



Modem manufacturers typically use popular RISC processors, such as ARM. The

common business model in such cases is to pay royalties per use of the core. By

avoiding the RISC processor completely, the modem manufacturer is significantly

saving on the cost of the modem, some amount of which can be passed on to the

consumer. Further cost saving results since the hardware requirement reduces to

the minimum essential. The end user now not only has a fancy new processor-based

PC, but is also able to save cost on peripherals like modem and exploit the

processor bandwidth efficiently.

Typical ADSL Modem design with entire modem as peripheral device Host CPU with Applications and ATM Stack

Modem Hardware Host CPU with Applications and ATM Stack Host CPU with Applications and ATM Stack

To Telephone Line

Host ADSL Modem with entire modem running on the host

Host CPU with Applications and ATM Stack

Modem Hardware Host CPU with Applications and ATM Stack Host CPU with Applications and ATM Stack

To Telephone Line

In order to differentiate the class of modems that run on general- purpose

CPU instead of the conventional embedded hardware modems, these are referred to

as host modems.

The Host System Advantage



The definition of a host-based system may be extended to cover even embedded

devices. In case of an embedded device, the host processor is typically the RISC

processor (e.g., StrongARM or Xscale) that controls the embedded device.

Let us consider the example of an embedded device. With the increasing

demands on wireless connectivity, it is useful to examine a present day pocket

PC. It is a notebook-sized device that needs to be carried around, hence is not

heavy. It, typically, has a high-end RISC core and a lot of hardware accessories

for voice band connection, wireless LAN, etc. These are typically peripherals

that can be attached. They do not form permanent features of the pocket PC,

since they not only contribute to the weight but also eat away the battery

power. For such mobile devices, miniaturization (for ease in transporting) and

long battery life are key design considerations. Both these requirements are

addressed by running significant or entire portion of the peripheral software on

the host RISC processor itself.

The power saving due to host-based systems is illustrated easily. The

following data points are from Intel Corp. The power consumption in mobile units

increased from 17 Watts during 1996 to 18.7 Watts during 1997. With addition of

mobile processors, increasing bus speeds, DVD players, etc, battery requirements

can only go up further. One of the most significant power drains in notebook

computers is the modem. Full DSP modems typically need about 1 Watt while

active.

This can be compared with PCTEL’s host signal processing voice band modems

that use roughly 10 percent of CPU cycles on a 266 MHz MMX processor, and

require about a quarter of the power. This reduction can contribute towards a

substantial increase in battery time for handhelds.

Host-based ADSL



The advantages of host-based design are demonstrated by a host-based ADSL

modem. ADSL is a technology that facilitates data rates of up to 8 Mbps to the

user on the existing telephone lines, with simultaneous use of telephone.

Typical modem design (see Figure 1) involves the modem designed as a PCI or USB

peripheral with a controller, data pump, and telephone line interface built on

the peripheral card. In the host-based ADSL design (see Figure 2), the entire

modem controller and the data pump is shifted to the host CPU and the hardware

just comprises the analog front-end and the telephone line interface.

In spite of the entire modem controller and data pump running on the CPU, the

modem needs only 35 percent of the CPU bandwidth on a Pentium III, 1GHz PC. With

increasing CPU speeds, the modem application occupies much less CPU bandwidth

permitting several other host-based applications to run simultaneously.

Due to reduction in hardware cost (no DSP, reduced form factor), the BoM

reduces by 30 percent as compared to complete embedded solution. Current DSL

chipset cost is approximately $15. The total BoM of the PCI DSL board is roughly

$25 (Data Source: TI). With 30 percent reduction in chipset price and further

reduction in board area and power supply requirements, host DSL hardware BoM is

likely to be $18 or less–a very significant cost reduction for the

large-volume PC OEM segment.

Host-based DSL modems are ideal DSL access solutions for home PC users. It is

expected that like the current default V.90 bundling with present generation

PCs, host DSL would be bundled with the next-generation PCs. They are also ideal

solutions for laptops and handhelds due to small form factor and reduced power

consumption.

The Host System

Design Challenge



Host-based system design is a paradigm shift in design methodology. It

involves system partitioning with more emphasis on running modules in soft form

and minimum or no hardware support. Architecting good host-based systems

requires a change in mindset for migrating from embedded architecture focus to a

flexible soft architecture focus.

The important question that needs to be addressed is how much of hardware

complexity should be traded with software complexity. Full soft implementation

will result in maximum hardware cost reduction, but may occupy too much of CPU

bandwidth to make perceptible difference to the user. Moreover, there is also

the concern that it may render running any other application infeasible.

On the other hand, unless majority of the complex hardware modules are

implemented in soft form on the host CPU, cost reduction may not be significant.

User behavior patterns on sets of applications that run together for a

particular device may influence the architecture decision too. There may be

several other challenges in architecting host-based systems, which will have to

considered carefully.

Addressing Users’ Concern



Another major challenge lies in user perception management. Users feel that the
CPU ‘will become very slow’ if host-based modems, codecs, and other

applications run on it. However, the fact is that most of the present

manufacturers run significant portions of voice band modem functions on the host

without announcing it. Users do not even notice it.

The flexibility offered by the host solutions can serve as a double-edged

sword. While remote debugging and upgrade features provide more options to the

user, they can also create more confusion and contribute to errors. It is

possible to provide user the control over CPU utilization by configuring the

application complexity (for example, reduced modem bit rate or reduced number of

frames per second for codec in order to reduce bandwidth occupied by

application). This increases confusion levels. The challenge is to develop

better user-friendly interfaces and educate the user.

Demands for increasing MIPS per dollar and MIPS per Watt open up new

opportunities for using the host CPU to implement functions previously performed

through DSPs or ASICs. Host solutions provide a distinctive advantage for client

devices, which are faced with increasing integration requirements while

maintaining/reducing power and cost. Architecting host-based solutions require

careful handling of usability aspects, besides considering trade-offs in terms

of functionality and cost.

Both to the manufacturer and the consumer, a host-based system means better

power management and smaller handheld devices, not to mention the fact that it

creates a justification forever increasing CPU speeds. It implies more

feature-rich products, remote upgrades, installation and debugging. With such a

win-win situation, it will not be a surprise at all that the products of the

future will be more and more host-based.

MT Arvind and G

Venkatesh Sasken Communication Technologies