Cyberclassroom : A Large-Scale Interactive
  Distance-Learning Platform
Lin，SW1,Cheng,K L2,Wang, R3,Zhou, H4 and Chan, S H5
1cs_lswaa@stu.ust.hk
2cs_cklaa@stu.ust.hk
3cs_wrx@stu.ust.hk
4cs_zhxaa@stu.ust.hk
5gchan@cs.ust.hk

Department of  Computer Science
The Hong Kong University of Science and Technology

ABSTRACT
Traditionally, lectures are held in classrooms where students are co-located in order to interact with the instructors. This model is not convenient or cost-effective for some students, e.g., those who are working, have access/mobility difficulties, live in remote areas, or experience some social isolation such as during the SARS period.

With the pervasive penetration of broadband Internet connections and wireless medium (Wi-Fi and 3G ),  we have developed a platform so that lectures can be conducted interactively over the Internet. The project, termed 'CyberClassroom ,' offers an experience similar to the traditional classroom today but with the students distributed in the Internet. Students anywhere may raise questions to their instructors at any time using their PCs or laptops over the Internet and wireless medium. Video, audio, and whiteboard are also streamed to the end-users in real time.  The system is scalable to hundreds of students and makes use of current off-the-shelf computing products and state-of-the-art networking techniques. It is cost-effective and effectively breaks the geographical limitations to attending lectures. Initial feedback from the students using the system is positive and encouraging.

Key words
Distance-learning, cyberclassroom, interactive, education, distributed,system, multimedia applications and communication networks

INTRODUCTION
With the tremendous growth in computing devices and computer usage, geographical distances between people have been significantly reduced. It is easy to name a few examples of Internet software that are currently in use: Net Meeting ,Internet telephony, ICQ, and so on.Using these,people in different pares of the world can communicate with each other more economically and efficiently than ever before.

Recently, the Hong Kong Government has been promoting and putting great effort into the idea of ' life-long learning '.However, working schedules and the pace of life in Hong Kong make it nearly impossible for people to learm at a fixed time in a fixed place. In the HKUST Computer Science Department，we have been developing a ' cyberclassroom ' platform for interactive distance-learning. This project aims at delivering live lectures to large groups of students. It allows students to participate in courses from any location at any time from any computer.  

We show in the following figure the system we developed. It consists of students distributed in the network. The professor may deliver his/her lectures via the Internet with audiovisual streams and Power Point. He/She may also interact with the students through polling.

There are several unique features and technologies in the project:
 Network module: We have designed and implemented a delivery mechanism so as to accommodate hundreds to thousands of students without the need for powerful servers and clients. Our network protocol has been presented in major conferences and seminars and has received wide interest.
Audio module: The audio has to be delivered to the students in good quality. We have used a highly compressed audio which does not take up much network bandwidth. The encoding standard we use is called G.723.
Video module: We have implemented the most recent encoding technology H.264, for video compression. It guarantees good quality of video while greatly minimizing the use of network and system resources.  As a result, bandwidth requirement for this software is greatly reduced.
Web interface: The focus of  this part is to support a web interface. With this feature, students can have lessons anywhere they can find a computer with access to the web. To further increase user-friendliness, the CyberClassroom project also provides installation files. Simple mouse clicks according to the guidelines make installation much more convenient. No compilation or linking, as required in last year's project, is necessary.
In the following, we explain the modules in detail.

NETWORK MODULE

Since the Cyberclassroom is a  real-time application and the number of people attending a class is usually quite large, a stable and scalable network layer is important. The main function of the network is to distribute the lecture materials, namely, video and audio, to all students attending the lecture. This technique is called multicasting.

Traditionally, there are two ways of multicasting.  Application Layer Multicast(ALM) and IP-multicast. In ALM, the multicast is done by a sequence of unicasts along a logical tree of nodes/users. It is used to solve the adaptability problem of IP-multicasting. Since the transmission of the data to a group of ALM hosts is done by unicast protocols such as TCP and UDP, it is supported by any router on the Internet. However, as all data transmissions are done by unicast, redundant transmission, which results in inefficient use of bandwidth, would be the major problem in using ALM. In contrast, by using the IP multicast technology, data only needs to be transmitted once and every host within the same group receive it. Compared with multiple unicast communication, IP multicast is more efficient as eliminating redundant transmission saves bandwidth. To utilize the power of both ALM technology and the IP-multicast technology,we therefore designed and implemented a new architecture named Island Multicasr(IM).

In Island Multicast, a group of classroom members namely a teacher and students, is divided into several smaller multicat-capable regions(or'islands'). All of these layer multicast(ALM) protocol, while within an island, IPmulticast is used. Island Multicast(IM) is a generic application-layer framework that works with any ALM protocol. For the Cyber Classroom project, we decide to use  ALMI again for inter island communication.

Using this structure, a menber in the session can be a ' parent node '  of other members, which need to forward all packets from a neighbor node to all other neighbors.  This peer-to-peer structure makes the system highly scalable in term of message multicasting. The major problem of this structure is that the data delivery depends to a large extent on the session ungracefully,then all of the downstream nodes will be disconnected from the main tree for a second. To solve this problem, we introduce some ' proxies ' to the system. The proxy should be highly reliable and set up by the IM network operator.  It is simply a member which serves many other members. The more proxies there are in the system, the more stable andreliable the network is..

AUDIO MODULE
Microsoft DirectX, a recognized standard for multimedia application development on the Windows platform, is used in the Cyber Classroom desktop version for operations on audio: recoding, compression, decompression, transmission and playing. (Fig 1 and Fig 2.).

Audio is transmitted in real-time so streaming technology was used in audio transmission. When one person speaks, all users should be able to hear. The audio transmission is one-to-many. Real-time audio is time-sensitive and requires low delay. Meanwhile teal-time audio can tolerate some loss. Therefore, the most suitable way to broadcast audio is using transport layer protocol UDP.

As bandwidth is the most valuable resource on the Internet, minimizing bandwidth by audio compression is one of the main aims of this application.  As this application is designed for hundreds of students using simultaneously, the required bandwidth is so large that the network cannot support it.  Therefore, audio compression is essential. In this  project we adapted G.723.1. G723.1 is a speech-compression algorithm standardized by ITU. G.723.1 has dul coding rates at 5.3 and 6.3 kbps. The vocoders process signals with 30 ms frames and low distortion. The input/output of this algorithm is 16 bit linear PCM samples. Middle bit rate G.723.1 vocoder delivers the highest compression ratio of  any of current ITU standards without compromising speech quality. Furthermore, the coding delay of our vocoder is very low (less than 0.5 second).

VIDEO MODULE
There are many different kinds of video compression standard, such as H.261, H263, and MPEG4, developed by several international institutes. In this project, the new video compression standard H264 standard provides a much larger compression rate and greatly reduces the use of network resources.    
The implementation work was based on software JM5.0 (the most updated version 7.6 ) of standard H264 implemented by ITU experts. To speed up the video compression,modification will be made based on the existing code of JM 5.0.
The process of speeding up the video codec started from investigating compitational bottlenecks by timer functions in<time.h>. From the figure below, Integer Pixel Inter-Prediction takes most of computational power.
To speed up the Integer Pixel Inter-Predictions. Inter-search mode other than 16×16 was disabled to ignore inefficient encoding processes in order to increase the encoding speed. Also, a new algorithm was applied for motion estimation.
 In many fast search algorithms for motion estimation, the Predictive Motion Vector Field Adaptive Search Technique（PMVFAST）algorithm, now improved based on the former MVFAST Algorithm, achieves best performance in terms of PSNR and speed-up ratio by reducing most of search points, These have very small probabilities that the current block finally reference to MMX technology was also applied to speed up various time-consuming calculations such as Sum of Absolute Difference (SAD) and YUV to RGB conversion,etc,
 
The original H.264 encoder and decoder work with file input and file output. In this project, the H264 encoder and decoder work with file input and file output. In this project, the H,264 codec must be modified so that it can work with web cam and network by providing necessary interfaces. As the JM software was implemented in C, it became necessary to build static libraries in order to achieve capability between C and C++, as well as to maintain the speed of the codec . The use of static libraries also allows easier modification and upgrading.

The following graph shows the speed optimization result during 4 stages of improvement.
Stage 1 ：Original
Stage 2 : Configuration Modification
Stage3 : PMVFAST
Stage 4: PMVFAST with MMX

Performance in terms of FPS for Video Codec (Tested in P4 2.4G Machine)

Beside the compression and decompression speed,video quality is also a great concern in the CyberClassroom project.The captured pictures below illustrate the testing of video quality:

The original foreman.qcif sequence(Left) and the decoded sequence (Right)

The original container.qcif sequence (Left) and the decoded sequence (Right)
As can be seen from the first figure , the face of  the foreman blurs because of the quantization performed at the encoder. For the same reason, the waves on the surface of the lake became less apparent . However, the frames are still clear in their entirety and their quality is acceptable for this project given the high compression rate.
WEB INTERFACE
It may happen that students may have to attend live lectures through public computers or computers with operating systems other than MS Windows. To tackle those problems, a web interface was developed. The web interface consists of a homepage ,a BBS system and a section for video and audio displaying.
The homepage was developed by Dreamweaver MX and Flash MX. Users can find all the general information of this project and the documentations.

Homepage
The BBS system is widely used in the Internet to make online discussions. All users can communicate with each other by posting and answering to the post. By using the free BBS system source developed by the 6k Group ,we built our own BBS system using ASP.

BBS
Several different implementations were considered for the video and audio part. But as our software uses specific video and audio standards (H. 264, G.723) and are thus difficult to write matching decoders in Java , we chose Windows Media Series Windows Media Encoder 9.0 as the video and audio encoder and the media stream can be easily decoded by the Windows Media Player.

Webpage with video player

Audio only
For a long time historically, the communication problems between different regions and countries were of great concern,especially for those where geographical barriers and distances made the traditional face-to-face delivery of instructions impossible.Modern technology and the Internet have revolutionized long-distance communications and thus the concept of distance learning has been able to be realized.
The CyberClassroom project provides great flexibilities for beaching:anyone who has access to the Internet can take live lessons remotely. Professors can conduct lectures as if in traditional classrooms, with minor changes utilizing functionalities such as PRS, whiteboard ,video and audio transmission.
CONCLUSIONS
Advantages of a CyberClassroom over traditional course delivery include virtually unlimited class size, and greater flexibility in terms of time and place to study. Last year, the outbreak of SARS resulted in the suspension of classes. With the ‘CyberClassroom’,students can attend lectures at home safely and effectively, and teaching sched ules need not be disrupted.
The Internet today consists of multicast ‘islands’ interconnected by multicast-incapable routers. In order to enable global multicast efficiently,multicast features should be used within an island while the islands are inter connected by unicast connections. In our project,we made use of an application layer framework called Island Multicast (IM),
Which organizes multicast delivery into two levels: at the upper level inter-island overlay is established while, at the lower intra-island level,IP multicast is used. Together with IP multicast, the scalability increased. This allows a larger pool of potential students and hence CyberClassroom offers advantages of virtually unlimited class size and greater choice over traditional course delivery.
In audio part, a compression algorithm G723.1 is used to convert speech into IP packets. G.723.1 delivers the highest compression ratio of any of the current ITU standards without compromising speech quality and retains very high voice quality,in spite of the huge compression.In practice,it is nearly impossible to detect a difference in voice quality between voices over IP. We can achieve a good audio quality with small delay (smaller than 0.5 second).
The CyberClassroom project implements H.264 standard for video encoding and decoding. It provides better video quality while reducing the video bit rate significantly. This promises a smooth video stream under limited bandwidth resources.
As regards installation,the use of Microsoft Installer(MSI) helped the project to provide with a new feature , which is not difficult to implement but very helpful . It reduced the unnecessary errors caused in compiling the source and setting up the software , and at the same time made our produce more user-friendly . For the web interface , several techniques and software packages were used , such as Macromedia Dreamweaver , in order to develop the web pages ; ASP programming to build the BBS system and Windows Media Encoder 9 Series to encode and broadcast the multimedia streams.
REFERENCES
K.-W.CHeuk, S.-H. Chan and J.Lee;Island Multicast: The Combination of IP Multicast with Application-Level Multicast. IEEE International Conference on Communication, to appear.
Pendarakis, Dimitrios, Sherlia Shi, Dinesh Verma, Marcel Waldvogel: ALMI: An Application Level Multicast Infrastructure, 2001
Tourapis, Alexis M., Oscar C. Au, Ming L. Liou: Predictive Motion Vector Field Adaptive Search Technique ( PMVFAST) Enhancing Block Based Motion Estimation.

译文
网络课堂：一个大型互动远程教学平台
林，SW1，程，K L2，王，R3，周，H4和H5陈，S
1cs_lswaa@stu.ust.hk
2cs_cklaa@stu.ust.hk
3cs_wrx@stu.ust.hk
4cs_zhxaa@stu.ust.hk
5gchan@cs.ust.hk

香港科技大学计算机科学系

摘要：
    传统上，为了与教员进行更好的交流互动，讲座一般在学生所处的教室进行。这种模式对于有些学生并不是很方便和划算，例如对于那些正在工作的、行动不便的、生活在偏远地区的，或者类似在SARS期间被社会隔离的人来说。
伴随着宽带互联网连接和无线媒体 （Wi-fi 和 3G） 的普遍渗透，我们开发了一个平台，让讲座可以通过互联网以交互方式进行。该项目被称为 “网络课堂”，提供了一种类似于现今的传统教室，但学生们分布在互联网的教学体验。学生可以在任何时间任何地点使用他们的 PC 或便携式计算机通过互联网和无线媒体向他们的导师提问题。视频、音频、以及白板也流向实时的用户终端。利用现有现成的计算产品和最先进的网络技术，该系统可扩展到数百名学生。它具有成本效益，能够有效地突破地域限制让学生来听取讲座。使用该系统的学生传来的最初反馈是积极和鼓舞人心的。

关键词：
远距离学习，网络课堂，互动、 教育、 分布式系统、 多媒体应用和通信网
引言
    随着计算设备和计算机使用的巨大增长，人与人之间的地理距离已显著减少。简单举几个目前正在使用的互联网软件的例子：网络会议，网络电话，ICQ聊天等等。利用这些，人们可以在世界的各个角落，与他人进行比以往任何时候都更经济、更有效地沟通。
近来，香港政府一直在努力推行“终身学习”的理念。然而，在香港的工作时间和生活节奏，让人们几乎不可能在一个固定的时间，固定的地点进行学习。在香港科技大学计算机科学系,我们已经开发了一个名为“网络课堂"的交互式远程学习平台。该项目旨在为广大学生提供现场讲座。它允许学生在任何时间任何地点通过任何一台计算机来参加课程。
我们在下图展示所开发的系统。它由学生的分布组成网络。教授可以通过互联网提供他/她的讲座录音和课件，也可以通过轮询与学生进行交流。
该平台有以下几个独特的功能和技术项目：
网络模块：我们设计并实现了一个传输机制，以便容纳成千上万没有强大的服务器的学生和客户的需要。我们的网络协议在最近召开的重大会议和研讨会上得以展示，并引起了广泛的关注。
    2、音频模块：在该模块中音频能够被很好地传递给学生。在传递过程中，我们使用了一个高度压缩的音频，并不会占用大量的网络带宽，使用的编码标准称为G.723。
3、视频模块：我们已经实现了最新的H.264视频压缩编码技术。它在保证了视频的高质量的同时，对于网络和系统资源的使用显著减少了，也因此，这个软件的带宽的要求大大降低。
4、网络接口：这一部分的重点是支持一个网络接口。有了这项功能，学生就可以在任何能找到一台计算机上网的地方学习。为了进一步提高用户友好性，网络课堂项目还提供安装文件。根据指导进行简单的鼠标点击使安装更方便。完全无需编译或链接，而这在去年的项目里是必要的。
下面，我们来详细解释一下模块。
1、网络模块
由于网络课堂属于实时应用，并且听课人的数量通常都是很大的，因此稳定的和可扩展的网络层显得极为重要。网络的主要功能是将演讲材料，即，视频和音频，分发给所有参加讲座的学生。这项技术被称为多播。
传统上，多播有两种方式：应用层组播（ALM）和IP组播。应用层组播，组播是沿节点/用户逻辑树序列的单播来做的。它是用来解决IP组播的适应性问题。由于数据向一组ALM主机的传输是通过如TCP和UDP之类的单播协议来完成的，它支持互联网上的任何路由器。然而，所有的数据传输是通过单播，冗余传输，从而导致带宽利用效率低下，这将成为层组播使用的主要问题。相比之下，利用IP组播技术，数据只需发送一次，同一组内的每个主机都能收到它。与多播通信相比，IP组播能够更有效的消除冗余传输，节省带宽。在ALM技术和IP组播技术的基础上，我们设计并实现了一个新的架构，将其命名为“岛组播（IM）”。
岛组播，即课堂上教师与学生，分成几个较小的多播能力的地区（或者“岛”）。当所有这些层组播（ALM）协议在同一个岛上，就使用IP 组播。岛组播（IM）是一个通用的应用层框架，它适用于任何层组播协议。对于网络课堂项目，我们决定再次使用ALMI为岛间通信。

                           岛组播
采用这种结构，会话中的成员可以是一个父节点的其他成员，这需要将所有的数据包从邻居节点传送到到其他所有的邻居。这种对等结构使系统在多播期也能保证信息的高效传播。这种结构的主要问题是数据传输在很大程度上取决于内的所有节点的可靠性。如果一个内节点突然断开，那么所有的下游节点也将在一秒钟内从主树中断开。为了解决这个问题，我们将“代理服务器”引入系统之中。代理服务器必须高度可靠，并且有IM网络运营商建立。这是一个简单的构件，提供许多其他成员。系统中的代理服务器越多，那么它所提供的网络就就越稳定，越可靠。

音频模块
     微软举，在Windows平台上进行多媒体应用程序开发的一个公认的标准，用于在网络课堂的桌面版操作：重新编码，音频压缩，解压缩，传输和播放。（图1和图2。）。
音频传输实时流媒体技术应用于音频传输。当一个人说，所有的用户应该能够听到。音频传输是一对多的。实时音频的时间敏感，需要低延迟。同时实时音频可以忍受一些损失。因此，播放音频的最合适的方法是使用传输层协议。

图1：音频流设计（教授方面）
图2：音频流设计（学生方面）
带宽是互联网上最有价值的资源，通过音频压缩带宽是该应用程序的主要目的。这个应用程序是专为数以百计的学生同时使用而开发，所需的带宽是如此之大，以致于网络无法支持它。因此，音频压缩是必不可少的。在这个项目中我们采用G.723.1。G.723.1语音压缩算法是一个标准的ITU。G.723.1在5.3和6.3千位/秒的DUL编码速率。该声码器信号处理30毫秒的帧和低失真。该算法的输入/输出是16位线性PCM样品。G.723.1语音声码器中的比特率提能够供任何当前ITU标准最高的压缩比，而不影响语音质量。此外，我们的声码器的编码延迟是非常低的（小于0.5秒）。
3、视频模块
    有许多不同类型的视频压缩编码标准，如H.261，H.263，MPEG4。他们是被一些国际机构开发的。在这个项目中，新的视频压缩标准H.264标准提供了一个更大的压缩率，大大降低了网络资源的利用。
  
比较不同的视频编码标准下比特率的PSNR
    工作的实施是基于专家设计的标准H264 ITU下的软件JM 5.0（最新版本7.6）。加快视频压缩，修改也将是基于JM 5.0现有的代码。
  
对原有的H.264编解码器各部分的时间消耗
    加快视频编解码器从在<时间定时器H>功能调查计算瓶颈的过程。从下面的数字，可以看出整像素的帧间预测需要的计算能力。
    加快整像素的帧间预测。为了提高编码速度，国际搜索模式禁用其他16×16忽略低效的编码处理。同时，一个新的算法应用于运动估计。
在许多快速搜索算法的运动估计,预测运动矢量场自适应搜索技术(PMVFAST)算法,现在基于前者MVFAST算法改进,实现最佳性能方面和加速比的PSNR值减少大多数搜索点,这些有非常小的概率,当前块最后引用MMX技术也被应用于加速各种耗时的计算如绝对差之和（SAD）和YUV到RGB的转换等。
原来的H.264编码器和解码器是使用文件输入和文件输出。在这个项目中，H264编码器和解码器是处理文件输入和文件输出。在这个项目中，H.264编解码器必须进行修改，使其能为网络摄像头和网络提供必要的接口。随着JM软件用C实现，它必须在C和C++之间建立静态库来保持的编解码速度。静态库的使用也可以更容易的修改和升级。
下面的图显示了在4阶段的改进速度的优化结果。
阶段1：原始
阶段2：配置修改
阶段3：运动
阶段4：运动技术

在压缩和解压缩速度，视频质量也在网络课堂项目的极大关注。拍摄的图片说明了视频质量的测试：

原foreman.qcif序列（左）和解码序列（右）

原container.qcif序列（左）和解码序列（右）
    可以从第一个看到第一个图，因为在编码器进行量化，工头的脸模糊。出于同样的原因，在湖的表面的波变得不明显。然而，该框架在整体上和他们的质量是可以接受的。这个项目给予高压缩率仍然是明确的。

4、网络接口
学生可以参加现场讲座通过公共计算机或计算机操作系统比Windows更可能发生。要解决这些问题，开发了一个Web界面。Web接口包含一个主页，BBS系统，部分音频和部分视频。
网页是由Dreamweaver MX，Flash MX开发的。用户可以找到所有的一般信息的有关项目和文件。

主页
    BBS系统是在互联网上广泛使用的在线讨论。所有的用户都可以通过发布和回答后互相沟通。通过提供免费BBS系统源码的6K集团，我们使用ASP建立了自己的BBS系统。
BBS
    几种不同的实现可以考虑使用视频和音频部分。但我们的软件使用特定的视频和音频标准（H. 264，G.723），因此很难用java写的匹配的解码器，我们选择了Windows Media系列Windows Media编码器9.0的视频和音频编码器和媒体流可以通过Windows媒体播放器轻松解码。

视频播放网页

只有音频
    很长一段时间过去了，沟通问题在不同地区和国家之间都受到极大关注，特别是那些在地理上的障碍和距离使得传统的面对面教授是不可能的。现代技术和互联网已经彻底改变了远程通讯，因此远程学习的概念已经能够实现。
    网络课堂提供了很大的灵活性：任何人谁有权访问互联网都可以生活在远程教育中。教授们可以进行讲座就像在传统的教室，有轻微的变化，利用功能如PRS，白板，视频和音频传输。

结语

    和传统的课程交付相比，网络课堂具有几乎无限的班级大小，以及在时间和地点上有更大的灵活性。去年，非典的爆发导致停课。“网络课堂”，学生可以安全的、有效的在家参加讲座，教学调度不需要中断。
    今天的因特网是组播的“岛屿”，不能由组播路由器互连。为了使全球多播组播功能有效，应在一个岛上使用组播功能，岛屿间的单播连接的。在我们的项目中，我们使用了一个应用层组播框架，称为岛组播（IM），其中组播为两层组织：在上层岛间覆盖建立的同时，IP组播技术在较低的水平内岛应用。连同IP组播可扩展性增加。这允许一个更大的潜在的学生，因此网络课堂提供传统教学的优势和几乎无限的可供选择的班级规模。
    在音频部分，压缩算法G.723.1是用来转换成IP数据包的演讲。G.723.1在提供比当前任何ITU标准都高的压缩比的前提下，尽管巨大的压缩，仍然保留非常高的声音质量。在实践中，它几乎不可能在不同IP间检测声音的语音质量之间的差异。我们可以实现延迟小（小于0.5秒）并且好的音频质量。
该网络课堂项目实现了H.264标准的视频编码和解码。它提供了更好的视频质量的同时大大减少了视频比特率。在有限的带宽资源下，给我们一个流畅的视频流。
至于微软的安装，安装使用（MSI）帮助项目提供了一个新的特征，这是不难实现，但很有帮助。它减少了在编译源和设置软件造成的不必要的误差，同时使我们的产品更加人性化。为了网络接口，一些技术和软件被使用，如Macromedia Dreamweaver是为了网页开发；ASP编程建立的BBS系统和Windows Media编码器9系列解码和播放多媒体流。

    参考文献

K. -w.cheuk，S. H. Chan和J.李；组播岛：IP组播和应用层组播相结合。IEEE国际会议上交流，出现。
pendarakis，Dimitrios，sherlia石，Dinesh Verma，马塞尔瓦尔德福格尔：ALMI：应用层组播的基础设施，2001
tourapis，亚历克西斯M.，奥斯卡C.金，明湖刘：预测运动矢量场自适应搜索技术（PMVFAST）提高基于运动估计块。