Thursday, July 15, 2010

How to Broadcast Multimedia Contents? VII Network Layer or Stream Layer Design

[How to Broadcast Multimedia Contents? I Introduction]
[How to Broadcast Multimedia Contents? II Lessons from The Channel]
[How to Broadcast Multimedia Contents? IV Hierarchical Modulation]
[How to Broadcast Multimedia Contents? V Overloaded Transmission and IC]
[How to Broadcast Multimedia Contents? VI Open-Loop MIMO for Broadcast Multicast Services]

Many broadcast/multicast infrastructures are engineered for delivering a wide range of contents, such as streaming media, multicast media and even IP datacast. Though all of them have the similar capabilities of delivering a pretty-much same set of services,  their bear technologies and network layer or stream layer designs are varied. In terms of physical layer, ATSC uses 8 VSB while DVB, FLO and ISDB use OFDM in their air interface designs. In terms of stream layer design, DVB-H uses IP, ISDB uses MPEG TS, T-DMB and FLO use their own mapping between application layer and MAC layer logic channels.


Tuesday, June 15, 2010

Fading Broadcast Channel Capacities: III Scalar Fading Channels

In wireless broadcast multicast services (BMS), a standard assumption is that each receiver knows something about the channel h, usually referred as channel side information (CSI) or channel quality information (CQI). This is a pretty reasonable assumption when the channel is fading slowly inside the design boundary since there are pilot symbols available for the receiver to estimate CQI. Since the channel and transmitted signals are independent to each other, the ergodic capacity of the fading channel with receiver side information is given by

Cfading( SNR, h ) = E log( 1 + |h|2 SNR ) ≤ CAWGN( E(|h|2) SNR ).

This means fading hurts or reduces the capacity in general if the transmitter knows nothing of the fading. This is different to the case that assumes the transmitter can estimate channel through CQI feedback and therefore can do some precoding on broadcast signals.

Since a statistic analysis on a log(*) probability function is non-trivial, one approach is to apply the well-know Maclaurin expansion on Cfading( SNR, h ) and obtain a polynomial series of it

Cfading( SNR, h ) = -ln-12 Σn=1[ (-SNR)nE(|h|n)/n ].

From here, it is much easy for us to find some interesting results in the following.

Low SNR Region

Cfading( SNR, h ) ≈ ln-12 SNR E(|h|) - ln-12 SNR2 E(|h|)2 K( h1/2)/2

where K() denotes Kurtosis function.
High SNR Region

Cfading( SNR, h ) ≈ E log( |h|2 SNR )

It is hard to evaluate the above in general since the random valuable h is inside the nonlinear function log( ). However, a simple close-form solution may be possible for some special cases, including Rayleigh channel model, Weibull channel model, and Nakagami-m channel model

Quantify Channel Fading

There are many literatures discussing how to quantify the amount of fading a channel may have. The parameter, Kurtosis, is one of them. Kurtosis essentially is the normalized fourth moment of a realvalued random valuable and indicates the ”peakedness” of a probability distribution. A high kurtosis means a large variance due to infrequent extreme deviations, which results in a sharp ”peak” and fat ”tails”. Another similar measurement is the amount of fading of a real-valued random valuable, which indicates the severity of channel fading in communications. The Kurtosis and amount of fading of major fading distributions are compiled in Table 1. A similar compilation can be found in [Shamai 01].

Table 1. The Kurtosis and amount of fading of various fading channel models

Saturday, April 24, 2010

Fading Broadcast Channel Capacities II: Gaussian Broadcast Channel

Instead of the traditional single-coverage model, a layered broadcast model with a two-layer coverage is considered here. In this model, the broadcast station (BS) broadcast two layers of signal to all mobile stations (MS) in the covered area. The signal for the inner coverage has the achievable rate of R1 and the achievable rate for the outer coverage is R2, where R1 > R2. The MS's located near to the outer coverage edge may only be able to reliably decode the data stream of a low rate R2 while the MS's close to the BS can decode both data streams with a high sum rate R1. There many ways for achieving this two-layer broadcasting, including frequency-division multiplexing (FDM), time-division multiplexing (TDM) and superposition precoding (SPC).

Figure 1. Achievable capacity region of broadcast channel. The coverage difference is 6dB.

One key aspect of studying two-layer broadcast is the finding of the achievable broadcast channel capacity, which states that a little throughput sacrifice on the users near to the coverage edge may lead to a big increase for the users with good reception [Cover 72]. This concept is illustrated in Figure 1, where the achievable rates of FDM, TDM and SPC are compared. It shows that SPC can outperform FDM and TDM most of the time. With moving up the network operation point, e.g., the single-coverage operation point (r2, r2), a little bit along the SPC curve a higher throughput r2+ \Delta > r2 is achievable. There are many methods implementing SPC. The most popular one is hierarchical modulation. More generally, SPC can be implemented by overloaded CDM.

Wednesday, April 14, 2010

Fading Broadcast Channel Capacities I: Introduction

Broadcast multicast service (BMS) has increasingly been popular for delivering multimedia content to mobile users. BMS can be implemented through either a dedicated digital broadcast infrastructure like DVB-T/H/S2, MediaFLO and DMB or a 3rd generation and beyond radio access network like UMTS or cdma2000 network. Traditional digital broadcast air interface and network are designed with the tradeoff between the achievable capacity and intended coverage in mind. The actual throughput is limited by the maximum transmit power and the worst channel condition so that each user in the coverage area can reliably receive services. Therefore, all covered users share services with same quality. The users under good reception condition may not have advantages, even though their achievable throughput can be much higher. In addition, there are also rising interests in upgrading existing digital broadcast systems with more services for new users while be able to keep existing users unchanged, delivering additional or better QoS's to users with advanced receivers while still be able to guarantee other users' services, and providing unequal protection on digital contents with high spectral efficiency [ DVB Project 00, FLO Forum 07, Jiang 05, 3GPP2 07]. This is also encouraged by the recent advance in scalable video coding with an extension of the H.264/MPEG-4 AVC video compression standard, which provides possible adaptation capabilities for the BMS applications with no feedback necessary. Many technologies are under investigation for these goals, e.g., rateless coding, superpositioning precoding (SPC), multiple-input multiple-output (MIMO) and selective retransmission. Backward compatibility and implementation complexity are among the major concerns in upgrading existing systems. Among the candidates, SPC, e.g., hierarchical modulation, is one of the promising technologies for upgrading existing systems with larger coverage and more QoS options while maintaining strictly backward compatibility.

In system design, BMS is traditionally taken as a special case of the well-known broadcast channel model, in which one transmitter serves multiple receivers and the transmitter either know or has no prior information of the channels between itself and each receiver. For BMS, usually it is assumed that the transmitter has no such prior chanel information. So far, it is well-known that maximum sum rate of Gaussian broadcast channel is achievable with SPC, with which two independent signals are superimposed at the transmit side and successive interference cancellation is employed at the receive side. SPC has been widely proven and included in various standards, such as DVB-T, FLO, UMB, etc., and is under study for DVB-H. Currently, SPC is implemented with hierarchical modulation. In UMB and FLO, e.g., two bit streams, named enhancement layer and base layer, are coded, bit-mapped and modulated a QPSK/QPSK symbol stream, which is transmitted over orthogonal-frequency division multiplexing (OFDM). In this case, the channel fading is thought to limit the achievable spectral efficiency of the broadcast channel and complicate the receiver design. Therefore, a good channel coding is usually employed for compensating the impairments by the channel fading.

The recent progress in multiuser receiver design and CDMA has brought new attentions on how to broadcast layered contents over broadcast channels [Verdu 99, Shamai 01, Wang 08]. From an information-theoretic perspective, the channel fading on CDMA signals isn't so bad as we thought [Shamai 01]. The achievable capacity of a CDMA multiuser fading channel may be higher than a single-user channel corrupted by the same fading distribution. In fact, it asymptotically approaches the single-user Gaussian channel capacity if the system load, which is defined as the number of symbols per chip, is high enough and a optimum receiver is used. On the other hand from a signal processing standpoint, the performance of successive interference cancellation (SIC) is asymptotically close to optimum receivers while the power imbalance between the desired signal and interference is large enough [Verdu 98]. Bringing these two observations together, a precoded OFDM was proposed to broadcast layer-coded content through fading channels [Wang 08] with superimposing an additional layer of Multi-Carrier Code-Division Multiplexing (MC-CDM) signal on top of the existing OFDM transmission. The power setting of the two layers of signals are specially controlled so that a SIC receiver can be employed at the receive side to separate the signals. Here, the proposed precoded OFDM is generalized as an overloaded Quasi-Orthogonal CDM (QO-CDM) and the impact of fading on broadcast channel with QO-CDM is analyzed.

Saturday, February 13, 2010

How to Broadcast Multimedia Contents? VI Open-Loop MIMO for Broadcast Multicast Services

What Is The Next for Mobile System Design? I A Single-Cell Model Perspective on Downlinks
[How to Broadcast Multimedia Contents? I Introduction]
[How to Broadcast Multimedia Contents? II Lessons from The Channel]
[How to Broadcast Multimedia Contents? IV Hierarchical Modulation]
[How to Broadcast Multimedia Contents? V Overloaded Transmission and IC]
[How to Broadcast Multimedia Contents? VII Network Layer or Steam Layer Design]

One most well-known space-time block coding (STBC) design is Alamouti code, which is the simplest open-loop orthogonal STBC. Alamouti code was designed for a two-transmit antenna system. It is a rate-1 code. It is the first open-loop encoding method with full diversity. Though orthogonal STBC has the advantages of relatively easy receiver design and full diversity, it is known that full-rate STBC don’t exist for more than 2 transmit antenna. From previous discussion, if two orthogobal STBCs are superimposed together and each of them experiences different channel fading, there would be multi-layer diversity in addition to potential superposition precoding gain. This means one STBC signal layer is in a bad channel condition, the other STBC signal layer may not. Therefore, the transmission of more than one layers may have higher achievable spectral efficiency.


Figure 1. A Quasi-Orthogonal Space Time Block Coding Example


One widely discussed example of quasi-orthogonal STBC is shown in Figure 1. From a receiver perspective, this quasi-orthogonal STBC obviously has a larger signal constellation size. In general, larger the signal constellation size is, high the spectral efficiency is achievable. Besides this, it may have so-called multilayer diversity when it experiences channel fading since each orthogonal STBC signal experiences different fading. However, there is no free lunch. The demodulation complexity may increase exponentially with the number of superimposed STBC layers.

Monday, February 1, 2010

How to Broadcast Multimedia Contents? V Overloaded Transmission and Interference Cancellation

[How to Broadcast Multimedia Contents? I Introduction]
[How to Broadcast Multimedia Contents? II Lessons from The Channel]
[How to Broadcast Multimedia Contents? IV Hierarchical Modulation]
[How to Broadcast Multimedia Contents? VI Open-Loop MIMO for BCMCS]
[How to Broadcast Multimedia Contents? VII Network Layer or Steam Layer Design]
[Precoded OFDM for BCMCS, 3GPP2 TSG-C NTAH C00-20080218-006R1]

Though hierarchical modulations have been widely adopted for enhancing broadcast multicast services, several issues are still left for future enhancements. The first consideration is the inter-layer interference (ILI) between layers. The ILI from enhancement layer(s) to base layer(s) is not additive white Gaussian. The base-layer achievable spectral efficiency is actually dented by ILI more than expected. In addition, for example, when orthogonal frequency division multiplexing (OFDM) is employed on the carrier, there is a frequency selectivity issue on the layered transmission in fading channels, especially when the channel bandwidth is far more than its coherent bandwidth. With the combination of hierarchical modulation and OFDM, the base layer signal and the enhancement layer signal experience the same channel fading. There is no multi-layer diversity, which can help boost the achievable throughput.

Figure 1. An example of overloaded OFDM transmission for upgrading existing OFDM multicast/broadcast

One simple overloaded transmission solution to upgrade existing OFDM based broadcast multicast traffic channel is shown in Figure 1. With this scheme, legacy mobiles can seamlessly operate in the upgraded network without additional change. The control overhead signal part is same. The pilot part is reused. Only the traffic channel part is upgraded. The new traffic channel part is layer-modulated and transmitted with an additional pre-coded OFDM modulated enhancement layer, where the symbols are precoded with Walsh-Hadmard matrix before OFDM. In an additive white Gaussian channel, this scheme has the superposition precoding (SPC) gain since it essentially is an implementation of SPC. However, the interference from the enhancement layer is randomized due to additional Walsh-Hadmard spreading. In the fading channel, additional multi-layer diversity gain is achievable, since the base layer and the enhancement layer are operating in different signal spaces. A general overloaded OFDM transmission structure is shown in Figure 2.

Figure 2. A general overloaded OFDM transmission structure

Figure 3. An example of strictly backward compatible upgrade of existing OFDM based broadcast multicast network

Another advantage of this upgrade architecture is there is little interference between legacy air interface (AI) and new AI. The base-layer signals of the neighboring legacy local operating infrastructure (LOI) and new LOI still are soft-combinable. The enhance-layer signals of the neighboring legacy LOI and new LOI are not overlapped to each other due to its limited overage. Legacy mobiles can properly decode the base-layer of both legacy LOI and new LOI with no problem. The only degradation is that legacy mobiles may not be able to decode the enhancement-layer of the new LOI.

From an information theoretic perspective, the schemes shown in Figure 1 and 2 are interesting. However, the questions remain, can it be implemented in realities or are the overloaded signals be reasonably easy to demodulate and decode? The answer is yes and an interference cancellation receiver should be employed. A successive interference cancellation (SIC) receiver can be enough to do the trick. On the other hand, SIC is one of the advanced receivers which are widely discussed and implemented in commercial products in realities. The reason can be explained in the following and in Figure 4.

Figure 4. Asymptotic multiuser efficiency (AME) of various multiuser receivers

From a receiver design standing point, the effective energy of user 1 e1 is always upper bounded by the actual energy A21. This typically is quantified through a parameter called multiuser efficiency or asymptotic multiuser efficiency (AME). Multiuser efficiency or ratio between the effective energy and actual energy, e1/A21, of an user, depends on the signature waveforms, received signal-to-noise ratio (SNR) and the employed detector and is always not larger than 1. The AME of several popular advanced receivers are plotted in Figure 4. In Figure 4, it shows for a low signaling loading factor β = K/N, the linear multiuser receivers like decorrelator and MMSE can achieve near-optimum spectral efficiency. However, for a high loading factor β, nonlinear multiuser receivers can help obtain the optimal spectral efficiency. The loading factor β is a parameter what percentage of the system degree of freedom are used by the transmitter. For a CDMA system, it usually denotes the ratio between the number of active users and the spreading gain. However, additionally when the received signal power imbalance between desired signal and interference is large, the performance of SIC is asymptotically close to optimum receivers. For a two-user case, the imbalance requirement is A2/A1 > (1- ρ2)/|ρ|. Fortunately this imbalance requirement also is the prerequisite to SPC transmission.

From the above discussion, it is easy for us to find that the combination of overloaded transmission and successive interference cancellation can be right ingredients to achieve fading Gaussian broadcast channel capacity.

Friday, January 22, 2010

What An Engineer Needs to Know About Patent Laws? I: Background

Intellectual Property Values

On August 9, 2000, CNN Money reported, a U.S. appeals court set a sooner-than-expected end to Eli Lilly and Co.'s reign as the sole marketer of Prozac, the popular antidepressant drug, a development that sent the pharmaceutical company's stock down by more than 30%. [“Eli Lilly gets Prozac blues “, CNN ] It is about $36 billion in Lilly stock value, roughly a third of the pharmaceutical giant’s then market capitalization. On June 7, 2002, New York Times reported that royalties from inventions earn an estimated $150 billion/year worldwide and are expected to grow 30% annually over the next 5 years. [“Trying to Cash in on Patents”, New York Times] From 1993 to 2003, IBM alone was reported to earn well over $10 Billion in revenue from licensing out its patents and be awarded over 22,000 patents, more than the 10 largest U.S. IT companies combined, including Intel, Microsoft, Sun, Dell and Apple. Meanwhile, Microsoft was reported to spend about $8 billion on R&D and filed about 3,000 patent applications each year in US. Obviously intellectual property is becoming a key asset of modern corporations. From a recent British survey, 40% of  a company's value is not shown in any way on its balance sheet. For some big corporations, including Walt Disney, Microsoft and P&G, more than 80% of their market value is said to be in intellectual property assets.

U.S. Patents granted, 1790–2008. Source: wikipedia.com  and uspto.gov

Patent Rights

In US Constitutional Authority, Article 1, Section 8, Clause 8, it states, “The Congress shall have Power ... to promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries”. The former US Presidents Thomas Jefferson said that “The issue of patents for new discoveries has given a spring to invention beyond my conception.” and Abraham Lincoln said, “The patent system added the fuel of interest to the fire of invention.” World Intellectual Property Organization states, “Under such regional systems, an applicant requests protection for the invention in one or more countries, and each country decides as to whether to offer patent protection within its borders.”

US Government grants a right to exclude, which means a US patent provides its owner with the legal right to prevent unauthorized making, using, selling, offering for sale in US and the importation into US, of the invention set forth and claimed in the patent. In exchange, the inventor must disclose how to make and use the invention. However,  it is Not a Right to Practice. It does not grant patent owners the right to practice the invention. ( e.g., government regulation may interfere ) From a business perspective, with the right to exclude, the granted patent rights may permit setting prices at a level not possible without patent protection and help erect barriers for entry into a market. They can also bring in more revenue through licensing or assignment.  On the other hand, if defensively used as part of a patent portfolio, they can be used to trade ( cross license ) rights to exclude, maintain product differentiation, develop reputation as innovator and help with credibility as well as advertising.