Considering the extent of an all-IP NGN (Next Generation Network) network, as well as fixed-mobile convergence, interworking between UMTS (Universal Mobile Telecommunication System) and external networks is crucial to provide end-to-end QoS (Quality of Service). The introduction of IMS (IP Multimedia Subsystem) as a control domain in the UMTS core network represents the evolution toward NGN network, and provides support for different multimedia services. Prominent advantage of UMTS is its ability to provide diverse services with QoS guarantees. Focus of this work is on the analysis of ensuring QoS for UMTS real-time traffic (Conversational and Streaming class) in a mixed network environment, composed of the UMTS core network and IP external domain. In order to achieve QoS for broadband multimedia real-time services, which have different requirements and higher restrictions in network parameters, the DiffServ (Differentiated Services) mechanism should be deployed as the QoS model. DiffServ architecture is based on a simple model where traffic entering a network is classified and conditioned at the boundaries of the network according to the DSCP (Differentiated Services Code Point) field in IP header and assigned to different behaviour aggregates. Within the core of the network, packets are forwarded according to the Per-Hop Behaviour (PHB) associated with the DSCP. PHB groups are implemented on network nodes using some packet-scheduling algorithms and AQM (Active Queue Management) mechanisms. The choice of a packet scheduler is important for providing QoS. When multiple queues are sharing common transmission media, there must be a scheduler to decide how to pick up packets from each queue to send out and is responsible for enforcing resource allocation to individual flows. PHB definitions do not specify any particular implementation mechanism and therefore the problem of PHB implementation gained significant attention. According to the 3GPP specifications, mapping between UMTS traffic classes and PHB groups can be done in GGSN (Gateway GPRS Support Node) element in UMTS core network. Standard QoS mapping between IP QoS classes and UMTS services enables both voice and video telephony (Conversational traffic class) to be mapped to the same QoS class and PHB group. Voice traffic has efficient and predictable statistical multiplexing charastreristics. However, video traffic when aggregated with voice, can strain this multiplexing performance because of its burstiness characteristic. As network resources are limited, video traffic with larger packet sizes than voice traffic can cause significant delay for voice packets on any congested segment of a network when both are aggregated together in the same QoS class. In the context of QoS assurance, the option of using the LLQ (Low Latency Queuing) algorithm on network elements with the objective of examining the mapping of voice and video telephony to different QoS classes (two different EF PHB groups) and virtual queues has been analysed. Within LLQ, PQ (Priority Queuing) is used for the scheduling of both voice and video telephony traffic with respect to other traffic classes. Voice and video telephony are mapped to two different DiffServ virtual queues and DiffServ PHB groups. Voice telephony is mapped to traditional EF (Expedited Forwarding) PHB group intented for criticial voice traffic and with decimal value of DSCP field 46. Video telephony is mapped to the second EF PHB traffic class referred to VOICE-ADMIT PHB with decimal value of DSCP field 44 and conforms with the traditional EF PHB group. It is also admitted using a CAC (Call Admission Control) procedure involving authentication, authorization and capacity admission. This differs from a real-time traffic class that conforms to the Expedited Forwarding PHB but is not subject to capacity admission. To verify the proposed idea, a simulation study was performed using NS-2 (Network Simulator Version 2) which is an event-driven simulator targeted at networking research and independent developed DiffServ4NS module for scheduling algorithms used in this paper. Changes were done regarding standard LLQ implementation in NS-2 which schedules packets only from one queue with highest priority, with the aim of scheduling two virtual queues (voice and video telephony) with PQ. This new concept was compared to other traffic scheduling algorithms like PQ (Priority Queuing), WFQ (Weighted Fair Queuing) and WRR (Weighted Round Robin). Two simulation experiments are conducted, which differ from each other according to localisation of congestion. In first experiment, congestion occurs due to only one bottleneck link overload while in second experiment, congestion occurs due to whole external IP network overload. The obtained results were statistically processed using SPSS (Statistical Package for the Social Sciences) version 17.0. The hypothesis proved that this approach can guarantee the upper boundary of IPTD (IP Transfer Delay), IPDV (IP Delay Variation) and IPLR (IP Loss Rate) for conversational and streaming traffic classes, while at the same time it does not completely exhaust bandwidth for interactive and background traffic classes in case of network congestion.