The North American cable industry's adoption of the Data-Over-Cable Service Interface Specification (DOCSIS) is solving long-standing problems of interoperability. But while this offers the cable industry the promise of lowered costs and good reliability, it presents only a partial solution to the vexing problem of interference on the upstream link between subscriber and headend.

Currently, an IEEE standards working group is developing a standard for an advanced, hybrid fiber/coax upstream physical layer called HI-PHY (High-Capacity PHY) for hybrid fiber/coax cable TV networks. HI-PHY is expected to bring a substantial increase in upstream throughput. To realize this potential, HI-PHY must deal with a multitude of noise sources present in the upstream transmission medium. To solve upstream interference problems, create a more robust physical layer and also provide higher throughput, three groups of potential solutions are emerging. This group includes: single-carrier modulation (quadrature amplitude modulation, or QAM); multi-carrier modulation (both discrete multi-tone or DMT and generalized DMT or GDMT); as well as spread spectrum (direct sequence code division multiple access or DS-CDMA). This article explores the relative merits of each of the approaches to the challenges of the cable TV upstream transmission medium.

Figure 1: Multi-band filter-bank communications system.

The common model used for analyzing modulation techniques is based on a multi-band filter bank communications system shown in Figure 1. A two-dimensional signal constellation is used to generate N parallel sub-symbols. These N sub-symbols make up a symbol. These symbols are generated at a rate of 1/T, where T is the symbol duration. Independent parallel sub-channels, each carrying a stream of sub-symbols, are distinguished by the index k. The stream of sub-symbols on different sub-channels are multiplexed orthogonally (different multiplexing methods are used with different modulation schemes) onto the transmission medium. At the receiving end, the signal is decoded into a stream of parallel sub-symbols. These sub-symbols correspond to (hopefully with high probability) the ones sent by the transmitter.

The model includes specially constructed multi-band filter banks at the transmitter and the receiver that essentially eliminate inter-symbol interference (ISI) and inter-channel interference (ICI) at the receiver.

Modulation techniques

It is assumed that all systems have the same data throughput, power and total signal bandwidth.

Single-carrier QAM-The basic system that uses time-division multiplexing to send N independent streams of sub-symbols serially onto the transmission medium.

Multi-carrier DMT-This method uses frequency-division multiplexing to modulate N independent streams of sub-symbols onto N independent sub-carriers, or sub-channels (sub-carriers are spaced at 1/T). A discrete Fourier Transform block can represent the multi-band filtering bank that performs this task. In this case, there is strict time isolation between symbols because of the Fourier Transform's inherent rectangular time windowing capability. This rectangular windowing function creates significant spectral overlap between sub-channels.

Multi-carrier GDMT-This method also uses frequency-division multiplexing to modulate symbols onto sub-channels, but with a difference. When comparing GDMT to DMT, the former does a better job of spectral isolation between sub-channels than DMT. This is accomplished by use of per sub-channel low-pass prototype filters that cover multiple symbol intervals. This approach contrasts sharply with DMT's emphasis on temporal isolation between symbols.

Spread spectrum DS-CDMA-This method uses code-division multiplexing to transmit N independent streams of sub-symbols by means of N orthogonal codes. In DS-CDMA, codes occupy all available bandwidth simultaneously.

White Gaussian noise

White Gaussian noise is the thermal background noise present in all electrical transmitting media. For all modulation systems under consideration, the signal/noise ratio (SNR) at the receiver's decision point is equivalent in the presence of White Gaussian noise. Therefore, no real advantage exists for any approach.

Narrowband interference: ingress

Approaches for suppressing ingress noise at the receiver vary with each modulation method.

Single-carrier QAM-With this form of modulation, ingress noise can be particularly troubling if no countermeasures are taken. However, if the bandwidth of the noise is small relative to the signal (the typical situation), a decision feedback noise-prediction method at the receiver can be an effective answer.

Multi-carrier DMT-Under this method, the sub-channels on or near where the interference is located will suffer severe degradation. However, multi-carrier DMT offers the option of optimally adapting bit- and power-loading for sub-channels at the transmitter, depending on the noise level, to remedy this problem. By using this alternative, performance can equal that of single-carrier QAM with ideal decision feedback prediction at the receiver.

Multi-carrier GDMT-With its low-pass prototype filters, this method provides better spectral isolation from sub-channels affected by ingress noise. In practice, the overall effective SNR of all sub-channels combined will be superior to DMT modulation with the same number of sub-channels. This performance gain is achieved at the expense of complexity introduced by pulse-shaping per sub-channel.

Spread-spectrum CDMA-With a matched, multi-band filter bank on the receiver, this method achieves the same performance as single-carrier QAM (with no decision-feedback noise prediction) where signal bandwidth, power and data throughput are equivalent. By using an equalizer on the receiver, this level of noise suppression can be significantly increased. However, this equalizer can also limit available orthogonal codes that can be used for transmission, thereby reducing system capacity. While rather complex multi-channel decision-feedback structures can eventually solve this challenge, this approach suffers from additional propagation errors.

Impulse noise

Single-carrier QAM-This method multiplexes and transmits sub-symbols serially in time onto the channel. As a result, impulse noise will affect the individual sub-symbols covered by the duration of the impulse noise event.

Multi-carrier DMT-For any particular impulse noise event, each multi-carrier DMT sub-channel will be affected equally (note that N sub-channels are affected). However, this impact is significantly less than the effect suffered by the individual sub-symbol used in the single-carrier QAM case. This is because of the much larger symbol duration for multi-carrier vs. single carrier systems for the same signal bandwidth.

Multi-carrier GDMT-This method also enjoys the same advantage as DMT in terms of longer signal duration. However, when the number of symbols hit is considered, this modulation technique suffers the most severe impact from impulse noise. This is because the pulse-shaping filter used for each sub-channel spans multiple symbol intervals. The other approaches avoid this problem because they isolate these symbols in time.

Spread-spectrum DS-CDMA-This method provides the same performance achieved by DMT. However, like DMT, all sub-channels are impacted equally by the interference.


Microreflections in the upstream transmission medium result from different types of impedance mismatches. The echoes that result from reflections are usually shorter than one symbol duration. Note that one symbol contains N sub-symbols.

For single-carrier QAM, channel imperfections such as microreflections cause the previously noted inter-symbol-interference (or ISI) between sub-symbols at the receiver. If ISI is not removed, the SNR at the decision point can suffer greatly. Generally, equalization at the receiver or transmitter is used to mitigate ISI. In the presence of microreflections, DS-CDMA provides an equivalent SNR performance at the decision point to single-carrier QAM (assuming equal bandwidth, power and data throughput). As with the single-carrier QAM example, DS-CDMA performance will receive a great boost if an equalizer is placed at either the transmitter or receiver.

For the same bandwidth, multi-carrier DMT and GDMT are more robust than single-carrier or spread-spectrum modulation techniques in handling microreflections. Multi-carrier modulation divides the transmission bandwidth into many narrower sub-channels. As a result, the microreflections result in nearly flat responses across these many narrower sub-channels. These require only a simple gain correction to mitigate these problems on each sub-channel.

Timing sensitivity

Timing sensitivity of modulation techniques is a particular concern in HFC networks. Here, timing sensitivity refers to any SNR variation at the decision point resulting from sampling the received symbol sequence outside the optimal time instant. Errors in ranging, synchronization, and/or time variations in the upstream channel may cause these timing errors.

The timing sensitivity of single-carrier QAM and DS-CDMA systems are theoretically equivalent as long as both are using the same bandwidth and data throughput. Multi-carrier DMT and GDMT modulation techniques are also sensitive to timing errors. However, by taking some countermeasures, multi-carrier systems can be made robust to timing errors by exploiting the greater symbol duration of these systems relative to single-carrier QAM.

Two steps can be taken to achieve this increased robustness. With DMT, a guard interval may be introduced between symbols. In the case of GDMT, a spectral guard-band is introduced between sub-channels. In both cases, the cost for this added robustness is paid for in lessened efficiency in data throughput for a given bandwidth.

End-to-end delays

In this article, end-to-end transmission delays refer to the total elapsed time between sending a digital symbol from the transmitter and receiving and decoding that particular symbol at the decision point of the receiver. This delay can also be referred to as PHY layer delay. The propagation delay introduced by the transmission channel is ignored because it is common to all techniques under consideration.

Single-carrier QAM modulation introduces the smallest end-to-end transmission delay. This is because of this method's special filter-bank structure which performs time-division multiplexing of sub-symbols. In this case, the delay corresponds to only several sub-symbol intervals, or a fraction of a symbol interval. In contrast, multi-carrier DMT and DS-CDMA modulations suffer transmission delays of at least a symbol interval in length. This is because of their block-loading of sub-symbols onto each symbol. For example, if there are N sub-channels in the DMT case, or N codes in the DS-CDMA case, then the end-to-end transmission delay suffered by these systems is generally N times greater than the delay suffered by single-carrier QAM.

With multi-carrier GDMT, the end-to-end delay is even more severe. This is a result of the pulse-shaping used to spectrally isolate sub-channels. In general, this delay will be an order of magnitude larger than the DMT or DS-CDMA case. This assumes equal bandwidth and data throughput for all systems under consideration.


Physical layer considerations are only meaningful in the context of the corresponding MAC layer, because the MAC and PHY layers are intrinsically related in cable TV networks. For example, the MAC layer has many control functions that are directly impacted by a modulation technique. This group includes upstream resource allocation, contention traffic and resolution, administration of PHY layer transmission profiles, allowable delays for various applications, software or microprocessor support of PHY layer control functions, PHY layer recovery times, etc. In addition, overall system requirements brought by a modulation technique need to be considered for evaluation of different modulation schemes.


For more information on this subject, consult J. Karaoguz, J. Yu and V. Eyuboglu, "Comparison of Single-Carrier, Multi-Carrier and Spread Spectrum Modulations for Upstream PHY layer in HFC CATV Networks," July 1998, IEEE 802.14a/98-018, San Diego.