VoIP & NGN Network Design with DiffServ QoS

Two Cisco Packet Tracer / Asterisk lab projects from my M.Sc. Communication Systems & Networks coursework at TH Köln, demonstrating SIP/RTP signalling analysis, VoIP service deployment, and DiffServ-based QoS for latency-sensitive traffic.

Stack: Cisco IOS (routers/switches), Asterisk, Wireshark, iPerf, SIP/RTP

---

Results at a Glance

QoS makes or breaks VoIP call quality once a link gets congested. These labs measured that effect directly rather than just asserting it.

1. DiffServ QoS impact on a congested 128 kbps link

Condition	VoIP (PCMA) throughput	Competing (iPerf) traffic	Outcome
Best Effort (no QoS)	10.11 kbps (7.9% of link)	117.89 kbps (92.1% of link)	VoIP starved — degraded audio
With DiffServ (EF/AF12 marking)	85.6 kbps (full allocation, guaranteed)	Capped at 41.72 kbps	Excellent QoE — no echo, minimal delay

DiffServ reserved $94.16\text{ kbps}$ total for VoIP ($85.6\text{ kbps}$ RTP media + $8.56\text{ kbps}$ SIP signalling, 10% overhead), leaving $33.84\text{ kbps}$ for best-effort traffic on the $128\text{ kbps}$ link.

2. Codec throughput comparison (measured via Wireshark I/O graphs)

Codec	Sampling rate	Application throughput	Data-link throughput
PCMA (G.711)	8000 Hz	64 kbps	85.6 kbps
G.722	16000 Hz	64 kbps	85.6 kbps
iLBC	8000 Hz	15.2 kbps	27.4 kbps

Note: iLBC trades audio quality for bandwidth resilience — it degrades less under congestion but sounds worse even when the link is clear.

3. Quality of Experience vs. bandwidth (measured delay, perceptual rating)

Codec	Link bandwidth	Delay	QoE rating
PCMA	128 kbps	0.75–0.95 s	10/10 — clear, minimal echo
PCMA	64 kbps	4.19–5.92 s	5/10 — noticeable echo
PCMA	32 kbps	4.41–9.7 s	Poor — unusable for real conversation
iLBC	128 kbps	0.71–0.90 s	8/10
iLBC	32 kbps	0.45–1.10 s	9/10 — best low-bandwidth option

4. SIP call setup latency

Measured directly from Wireshark capture (time between first INVITE and 180 Ringing):

$$\text{Call setup latency} = 23.0500 - 11.35565 = 11.69435\text{ ms}$$

---

What Was Built

Lab A — VoIP-enabled enterprise LAN with DiffServ QoS

• Topology & Routing: 3-LAN topology (Asterisk server, VoIP hardphones/softphones, routers, switches) with NAT and static routing.
• SIP Trunking: Configured for cross-network call routing.
• DiffServ Domain: VoIP media classified as Expedited Forwarding (DSCP 46), SIP signalling as Assured Forwarding AF12 (DSCP 12), and everything else as Best Effort (DSCP 0).
• Policy Rules: ACL-based traffic classification + policy maps for ingress DSCP marking and egress bandwidth allocation.
• Stress Testing: Validated under a constrained $128\text{ kbps}$ serial link with parallel iPerf-generated congestion.
• Full report with topology diagrams & Cisco config

Lab B — Next-Generation Network: SIP proxy/peer-to-peer, NAT traversal, multi-domain interconnect

• Architecture: VoIP service provider network with Asterisk, hard/softphones, NAT, and static routing.
• Protocol Analysis: SIP registration and authentication (challenge-response) analysis via Wireshark.
• Media Flows: Peer-to-peer RTP vs. proxy-mode RTP comparison, including NAT traversal behavior (RFC 3581 symmetric response routing).
• NAT Mapping Dependencies: Demonstrated that calls failed when initiated from outside a NAT'd network (empty translation table) and succeeded when initiated from inside it.
• Trunk Interconnection: Cross-team SIP trunk interconnection with live transcoding between PCMA and iLBC codecs.
• Full report with call-flow diagrams & Cisco config

---

Key Takeaways

1. Traffic Marking Matters: DSCP-based traffic marking and policy-map bandwidth reservation reliably protect VoIP quality even when a link is experiencing over 92% congestion from competing traffic.
2. Codec Selection: Codec choice is a real engineering tradeoff. PCMA/G.722 sound best but need roughly $3\times$ the bandwidth of iLBC and degrade heavily under congestion; iLBC is the right design choice for severely constrained links.
3. NAT Asymmetry: NAT silently breaks inbound VoIP unless a translation entry already exists, a call originates from behind the NAT, or explicit port forwarding / SBC-style handling is configured.
4. Signaling vs. Media Split: SIP signalling (via rport/RFC 3581) can survive NAT independently of RTP media, which is a common real-world cause of the "call connects but there is no audio" symptom.

---

Repository Files

• diffserv-qos-report.pdf: Full conference-style writeup of Lab A (topology, DSCP config, throughput tables, Wireshark I/O graphs).
• ngn-team1-report.pdf: Full writeup of Lab B (NAT, peer-to-peer SIP, transcoding, cross-team interconnect).
• voip_qos_diffserv_capture.pcapng: The raw Wireshark network capture file containing the SIP signaling exchanges and RTP audio media streams used in our QoS analysis.
• lab3_VoIP_iperf_Qos.jpg: Performance visualization displaying the iperf throughput metrics and network behavior during active traffic congestion testing.

---

Coursework completed as part of M.Sc. Communication Systems & Networks, TH Köln, under Prof. Dr. A. Grebe.